(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Biodiversity Heritage Library | Children's Library | Additional Collections
Search: Advanced Search
Anonymous User (login or join us)
Upload
See other formats

Full text of "Updating the geographic practice cost index : the practice expense GPCI : final report"

! 



a gfcgj — — - 

REPORT DOCUMtWIATWy 

PAQE- 



PB-9A-161098 



4. TWa 



Updating the Geographic Practice Cost Index: 
The Practice Expense GPCI 

FINAL REPORT and APPENDICES TO FINAL REPORT 



?fay 1994 fPevised) 



Gregory C. Pope; Debra A. Dayhoff; John E. Schneider; 
Angela B MamJL] and Killard W. Adamache; Stephen Zuckermdn 



7. 



Health Economics Research, Inc. 
300 Fifth Avenue 
6th Fldor 

Waltham, MA 02154 



<° SOO-89-0050.5 
na 



Health Care Financing Administration 
2-B-14 Oak Meadows 31dg. 
6325 Securitv Blvd. 
Baltimore, MD 21207 



U. Typaa* 



: inal 



14. 



IS. Sua 



Sea also PB94-161072; PB94-161080; and PB94-161106. This study supports the Notice of Proposed Rule Making 
(BPD-789-P) "Medicare Program Refinements to the Physician Fee Schedule Geographic Factor Value* and Other 
Changes", forthcoming in the Federal Register In June 1994. 



20= 



(Unite 200 

The Omnibus Budget Reconciliation Act (OBRA) of 1989 reformed Medicare payment for physician service* as of 
January 1, 1992. Cn that date. Medicare began paying physicians on the basis of a national Medicare Faa 
Schedule (MFS) Instead of usual, customary, and reasonable charges. Payments to physicians are now 
calculated from national uniform relative values and various conversion factors adjusted for justifiable 
geographic differences in physicians' cost of practice using a geographic adjustment factor termed tne 
Medicare Geographic Practice Cost Index (GPCI). Subsequently, OBRA 1990 required that the GPCI be reviewed 
periodically and revised as necessary. The subject study supported HCFA's review of the original GPCIs; 
assessed the Medicare practice expense GPCI (PEGPCI) In effect from 1992-1994; and recommended a number of 
alternatives that might be used to update the PEGPCI. The report consists of three parts — Pare I the 
employee wage index. Part II the office rental index, and Part III supplies, equipment, and miscellaneous 
expenses. Study recommendations for the wage Index component are to use median hourly 1990 Census earnings 
for selected occupations, for counties within large metropolitan areas. For the office rental index use the 
FY 1994 Housing and Urban Development Fair MarKet Rent values, modified by county-specific rather than 
metropolitan rents in consolidated metropolitan areas. F-r miscellaneous, equipment, and supplies no new 
data sources were found, thus, the current policy of no geographic variation was ratified. Relative value 
(RV) weighting and rescaling for Medicare budget neutrality (factor 00125) were incorporated into the 
final values. The extensive AppenaixeB include IV-1 the population versuB RV weighting analysis; IV-2 the 
1996 PEGPCI, tr.e 1996 rescaled PEGPCI, the 1995 rescaled PEGPCI, the 1992 (1992-1994) current PEGPCI by 
Medicare payment locality; IV- 3 ranKs the Medicare payment localities in order of descending difference 
oetween the 1996 rescaled PEGPCI anc the 1992 (current) PEGPCI. 



17. Document Analysis a. 



b. ldenttAars/Ope>v£ndad Term* 

Medicare physician payment; practice expense: Geographic Practice Cost Index; PEGPCI; Laspeyres input price 
i.-.dex; Medicare .-ee Schedule: MFS; Gaccrapnic Adjustment Factor; GAF; employee wage index; office rent 
-r.dex; and supplies, equipment, ana miscellaneous prices. 



c. COSATI Flald/Qrouo 



It. Availability St ate m aw ; 

R elease IJnl: 



.rj:tea 



IS. Security Class (This Raoort) 

'ry r]_\^c :TT-TT-n 



2U No. of Pa«es , 



20. Security Class (This Pace) 

?r\*CLASSI r TEr> 



22. Pries 



5 3 dspj£iiM*>u*j* 



(Sm ANSU239.1S) 



See tnsi 



art Osa era e 



OPTTONA1. rom 272 (4-77) 

rr ill MTIt-IH 



hE3 



HEALTH ECONOMICS RESEARCH, INC. 



30C Rf r i~i Avenue. 6>h 
Waithsm, VIA 02154 



'6 X 7) -'.37-C2C2 Fa% 



UPDATING THE GEOGRAPHIC 
PRACTICE COST INDEX: 

THE PRACTICE EXPENSE GPCI 

Final Report 



Submitted by : 



Gregory C. Pope, M.S. 
Debra A. Dayhoff, Ph.D. 
Killard W. Adamache, Ph.D. 
John E. Schneider, M.A. 
Angela R. Merrill, B.A. 

Health Economics Research, Inc. 



and 



Stephen Zuckerman, Ph.D. 

The Urban Institute 
2100 M Street, N.W. 
Washington, DC 20037 




May 1994 



This report was prepared under Contract No. 500-89-0050 from the Health Care Financing Administration to Health Economics 
Research, Inc. The assistance of our HCFA Project Officer, Sherry Terrell, PhD., is gratefully acknowledged. The statements 
contained in this report are solely those of the authors and do not necessarily reflect the views or policies of the Health Care 
Financing Administration. Trie contractor assumes responsibility for the accuracy and completeness of the information 
contained in this report. 



TABLE OF CONTENTS 



PAGE 



EXECUTIVE SUMMARY . . ES-1 

INTRODUCTION 1-1 

PART L EMPLOYEE WAGE INDEX 

1.0 INTRODUCTION 1-1-1 

1.1 Background on the Employee Wage Index 1-1-1 

1.2 Criticisms of the Employee Wage Index . , 1-1-1 

1.3 Overview of Part I of the Report 1-1-2 

2.0 DATA FOR THE EMPLOYEE WAGE INDEX 1-2-1 

2.1 Description of Census Data 1-2-1 

2.2 Sample Sizes of Census Data 1-2-1 

3.0 METHODS FOR COMPUTING NON-PHYSICIAN EMPLOYEE WAGE 1-3-1 

3.1 Occupation Shares for Employee Occupations 1-3-1 

3.1.1 Data Source 1-3-1 

3.1.2 Computation 1-3-2 

3.1.3 Comparison with Current Shares 1-3-2 

3.1.4 Validation Versus AMA SMS Shares 1-3-4 

3.2 Non-Physician Employee Wage Indexes 1-3-4 

3.2.1 Create County File of Wages for Occupation Categories 1-3-4 

3.2.2 Calculate County Index 1-3-6 

3.2.3 Translate County Index to Medicare Payment Localities 1-3-6 

4.0 ALTERNATIVE NON-PHYSICIAN EMPLOYEE WAGE INDEXES 1-4-1 

4.1 Updating With 1990 Census Data 1-4-1 

4.2 Non-Manufacturing Administrative Support Staff Wages 1-4-4 

4.3 County-Specific Wages in CMSAs 1-4-6 

4.4 Mean Earnings 1-4-8 

4.5 Comparison to the PPS Hospital Wage Index 1-4-11 

5.0 FINAL EMPLOYEE WAGE INDICES 1-5-1 

PART IL OFFICE RENTAL INDEX 

1.0 INTRODUCTION II-l-l 

1.1 Background on the Office Rental Index II-l-l 

1.2 Criticisms of the Rental Index II-l-l 

1.3 Overview of Part II of the Report II-1-2 

2.0 REVIEW OF DATA SOURCES FOR OFFICE RENTAL INDEX II-2-1 

2.1 Previous Data Searches II-2-1 

2.2 Review of Private Sources of Commercial Office Rents II-2-2 

2.2.1 Building Owners and Managers Association II-2-2 

2.2.2 Institute of Real Estate Management II-2-3 

2.2.3 Society of Industrial and Office Realtors II-2-5 



TABLE OF CONTENTS (continued) 



PAGE 



2.2.4 The Urban Land Institute II-2-5 

2.2.5 Limitations of Private Commercial Rental Data II-2-5 

2.3 Internal Revenue Service Tax Returns II-2-7 

2.4 General Services Administration Rental Survey II-2-8 

2.5 U.S. Post Office Rental Data II-2-9 

2.6 Department of Housing and Urban Development (HUD) Fair Market 

Rents (FMRs) II-2-10 

2.6.1 General Description II-2-10 

2.6.2 Strengths of the FMRs for the GPCI Rental Index II-2-11 

2.6.3 Weaknesses of the FMRs for the GPCI II-2-11 

2.6.4 Effect of Residential Rent Control on the FMRs II-2-13 

2.7 Primary Data Collection II-2-14 

2.8 Conclusions on Data for the Office Rental Index II-2-16 

3.0 ALTERNATIVE OFFICE RENTAL INDICES H-3-1 

3.1 Rationale for Office Rental Index Alternatives II-3-1 

3.1.1 Most Recent Single Year II-3-1 

3.1.2 Three Year Average II-3-1 

3.1.3 County-Specific Rents in CMSAs II-3-2 

3.1.4 Two Versus Four Bedroom FMRs II-3-2 

3.2 Methods Used to Calculate Alternative Office Rental Indices II-3-3 

3.2.1 Create County File of FMRs II-3-3 

3.2.2 Replace New York City FMR II-3-4 

3.2.3 Calculate County FMR Index II-3-4 

3.2.4 Translate County Index to Medicare Payment Localities II-3-5 

3.2.5 Create County-Specific Rents in CMSAs II-3-5 

3.3 Changes in the Office Rental Index From Updating II-3-6 

3.3.1 Single Most Recent Year: FY 1993 FMRs II-3-6 

3.3.2 Three Year Average: FY 1993, 1992, 1991 FMRs II-3-9 

3.3.3 County-Specific Rents in CMSAs II-3-9 

3.3.4 Two Bedroom Versus Four Bedroom FMRs II-3-14 

4.0 NEW YORK CITY AREA OFFICE RENTAL INDICES II-4-1 

4.1 Evidence From Physician Surveys II-4-2 

4.2 Apartment Rents by County in the New York Metropolitan Area II-4-4 

4.3 Office Rental Indices for New York City Area Localities II-4-4 

5.0 VALIDATION OF OFFICE RENTAL INDEX USING PHYSICIAN SURVEY 

DATA H-5-1 

5.1 Previous Validations H-5-1 

5.2 Validation using the PPCIS II-5-3 

5.2.1 Data Description II-5-3 

5.2.2 Descriptive Results II-5-4 

5.2.3 Regression Results II-5-7 

5.2.4 Variation in Costs After Adjusting for the GPCI II-5-12 

5.2.5 Conclusion IJ-5-12 

6.0 FINAL OFFICE RENTAL INDICES H-6-1 



REFERENCES 



TABLE OF CONTENTS (continued) 



PAGE 



PART ID: SUPPLIES, EQUIPMENT, AND MISCELLANEOUS PRICES 



1.0 INTRODUCTION ffl-1-1 

1.1 Overview of Part III of the Report III-l-l 

2.0 INPUT PRICE DATA ffl-2-1 

2.1 Secondary Data III-2-1 

2.1.1 Review of Previous Searches for Secondary Data III-2-2 

2.1.2 Results of HER Search Efforts III-2-4 

2.1.3 Conclusions III-2-9 

2.2 Primary Data Collection III-2-10 

2.2.1 Design Issues for a Survey of Input Prices III-2-10 

2.2.2 Types of Primary Surveys III-2-12 

2.3 Concluding Comments III-2-12 

3.0 ANALYSIS OF PHYSICIAN SURVEY DATA ffl-3-1 

3.1 Review of Previous Studies III-3-1 

3.1.1 AMA Study III-3-1 

3.1.2 Urban Institute Study III-3-2 

3.1.3 Limitations of Using Expenditure Data to Infer Input Price 
Variation III-3-3 

3.2 Analysis of PPCIS Practice Costs and Equipment Prices III-3-5 

3.2.1 Description of the 1988 Physicians' Practice Costs and Income 
Survey III-3-5 

3.2.2 Methods III-3-6 

3.2.3 Geographic Variation in Practice Costs and Equipment Prices . . III-3-8 

3.2.4 Correlation with Other GPCI Input Price Measures III-3-15 

3.2.5 Conclusions III-3-17 

4.0 CONCLUSIONS FOR POLICY ffl-4-1 

4.1 Effect of Hypothetical Supplies, Equipment, and Miscellaneous Price 
Variation on the Geographic Adjustment Factor (GAF) III-4-1 

4.2 Policy Options III-4-3 

4.3 Primary Data Collection in the Long Run III-4-4 

REFERENCES 

PART IV: THE PRACTICE EXPENSE GPCI IV-1 



TABLE OF TABLES 



PAGE 



PARTI: EMPLOYEE WAGE INDEX 



TABLE 2-1 DISTRIBUTION OF 1990 CENSUS SAMPLE SIZES FOR 

EMPLOYEE OCCUPATION GROUPS, BY MEDICARE 
PAYMENT LOCALITY 



1-2-2 



TABLE 3-1 OCCUPATIONAL EXPENDITURE SHARES FOR THE 

EMPLOYEE WAGE INDEX 



1-2-3 



TABLE 4-1 MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE 

CHANGES IN THE EMPLOYEE WAGE INDEX AS A RESULT 
OF UPDATING WITH 1990 CENSUS DATA 



1-4-2 



TABLE 4-2 MEDICARE PAYMENT LOCALITIES EXPERIENCING THE 

LARGEST CHANGES IN THE EMPLOYEE WAGE INDEX AS A 
RESULT OF USING WAGES OF NON-MANUFACTURING 
INDUSTRY CLERICAL WORKERS 



1-4-5 



TABLE 4-3 MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE 

CHANGES IN THE EMPLOYEE WAGE INDEX AS A RESULT 
OF USING COUNTY LEVEL DATA WITHIN CMSAS 



1-4-7 



TABLE 4-4 MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE 

CHANGES IN THE EMPLOYEE WAGE INDEX AS A RESULT 
OF USING MEAN RATHER THAN MEDIAN EARNINGS 



1-4-9 



TABLE 4-5 MEDICARE PAYMENT LOCALITIES WITH LARGE 

DIFFERENCES BETWEEN THE GPCI EMPLOYEE WAGE 
INDEX AND THE PPS WAGE INDEX 



1-4-13 



PART IL OFFICE RENTAL INDEX 



TABLE 2-1 MEAN OFFICE RENT IN DOLLARS PER SQUARE FOOT, 

1991: SELECTED AREAS 



II-2-4 



TABLE 3-1 LOCALITIES EXPERIENCING LA RGE CHANGES IN OFFICE 

RENTAL INDEX AS A RESULT OF USING 1993 FAIR MARKET 
RENTS (FMRs) 

TABLE 3-2 LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE 

RENTAL INDEX AS A RESULT OF USING A BLEND OF 1987 
AND 1993 FAIR MARKET RENTS (FMRs) 

TABLE 3-3 LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE 

RENTAL INDEX AS A RESULT OF USING A BLEND OF 1991, 
1992, AND 1993 FAIR MARKET RENTS (FMRs) 



II-3-7 



II-3-10 



II-3-11 



TABLE OF TABLES (continued) 



TABLE 3-4 LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE 

RENTAL INDEX AS A RESULT OF USING COUNTY-SPECIFIC 
RENTS, 1993 



PAGE 



II-3-13 



TABLE 3-5 LOCALITIES WITH THE LARGEST DIFFERENCES BETWEEN 

OFFICE RENTAL INDICES BASED ON TWO- AND 
FOUR-BEDROOM FAIR MARKET RENTS (FMRs) 

TABLE 4-1 NEW YORK CITY AREA MEDIAN PHYSICIAN OFFICE COST 

PER SQUARE FOOT: PHYSICIAN PRACTICE COST AND 
INCOME SURVEY, 1988 

TABLE 4-2 NEW YORK CITY AREA MEDI <\N PHYSICIAN OFFICE 

EXPENSES PER SQUARE FOOT: AMA DATA, 1991, 1990 

TABLE 4-3 COUNTY-SPECIFIC APARTMENT RENTS FOR THE NEW YORK 

CITY PMSA 



II-3-15 



II-4-3 



II-4-5 



II-4-6 



TABLE 4-4 OPTIONS FOR UPDATED NEW YORK CITY AREA OFFICE 

RENTAL INDICES 



II-4-8 



TABLE 5-1 COMPARISON OF PHYSICIAN OFFICE EXPENSE PER SQUARE 

FOOT TO FAIR MARKET APARTMENT RENTS 



II-5-6 



TABLE 5-2 REGRESSIONS OF PPCIS COST PER SQUARE FOOT ON GPCI 

OFFICE RENTAL INDEX 



II-5-9 



TABLE 5-3 ELASTICITY ESTIMATES FOR SUBGROUPS FROM 

REGRESSION OF PPCIS EXPENSE PER SQUARE FOOT ON 
GPCI INDEX OFFICE RENTAL 



II-5-10 



TABLE 5-4 TESTS FOR VARIATION IN GPCI ADJUSTED EXPENSE PER 

SQUARE FOOT 



II-5-13 



PART ffl: SUPPLIES, EQUIPMENT, AND MISCELLANEOUS PRICES 

TABLE 3-1 NUMBER OF VALID OBSERVATIONS IN PPCIS EQUIPMENT 

SUPPLEMENT 



III-3-7 



TABLE 3-2 MEDIAN EQUIPMENT PURCHASE PRICES BY CENSUS REGION 

AND URBANICITY III-3-9 

TABLE 3-3 PRACTICE EXPENSE REGRESSIONS III-3-12 

TABLE 3-4 ELECTROCARDIOLOGY EQUIPMENT (EKG) PRICE 

REGRESSIONS III-3-13 



TABLE OF TABLES (continued) 



PAGE 

TABLE 3-5 F TESTS ON GEOGRAPHIC VARIABLES IN EXPENSE / PRICE 

REGRESSIONS III-3-14 

TABLE 3-6 ELASTICITIES OF GPCI PRICE INDICES IN EXPENSE/ PRICE 

REGRESSIONS III-3-16 

TABLE 4-1 EFFECTS ON GAF OF HYPOTHETICAL VARIATION IN 

SUPPLIES, EQUIPMENT, MISCELLANEOUS INPUT PRICES III-4-2 



EXECUTIVE SUMMARY 



ES1.0 THE PRACTICE EXPENSE GPQ 

The Geographic Practice Cost Index (GPCI) is an input price index used to adjust 
Medicare Fee Schedule payments for geographic variations in pnysicians' practice costs. The 
GPCI has three components: physician work, practice expense, and malpractice insurance. 
The practice expense GPCI, in turn, has three components: 

1. nonphysician employee wages; 

2. office rent; and 

3. supplies, equipment, and miscellaneous inputs. 

The Health Care Financing Administration contracted with Health Economics Research, Inc. 
to update the GPCI. This report analyzes alternatives for updating the current (1992) practice 
expense GPCI with 1990 Census and other data. Companion reports discuss updating the 
physician work and malpractice GPCIs. In this Executive Summary, we first discuss 
updating the cost shares used to combine the three components of the practice expense GPCI, 
then we discuss updating each of the component indices, and finally, the effect of updating 
on the entire index. 

ES2.0 PRACTICE EXPENSE COST SHARES 

The three component indices of the practice expense GPCI are combined into an 
overall index using the average shares of each of the three cost categories—employee wages; 
office rent; and supplies, equipment, and miscellaneous—in physician practice expenses. Cost 
shares for the updated (1996) GPCI are discussed in detail in a companion report (Dayhoff, 
Schneider, and Pope, 1994). There it is concluded that the GPCI should utilize the same cost 
shares as the most recent Medicare Economic Index (MEI), which are derived from 1989 
American Medical Association Socioeconomic Monitoring System survey data on physician 
practices. The MEI cost shares for practice expense are: 39.8% for employee wages, 25.1% 



gpci\ pracexp\execsum 



ES-1 



for office rent, and 35.1% for supplies, equipment, and miscellaneous. 1 Thus, the 1996 
practice expense GPCI (GPCIPE96) is computed as: 



GPCIPE96 = 0.398*EMPWAGE + 0.251*OFFRENT + 0.351*SEM, 



where 



EMPWAGE = index of nonphysician employee wages; 
OFFRENT = index of office rental prices; and 

SEM = index of supplies, equipment, and miscellaneous prices. 

ES3.0 EMPLOYEE WAGE INDEX 

The practice expense GPCI employee wage index measures the relative wages in 
different areas of nonphysician employees in physician offices. The 1992 GPCI employee 
wage index was computed from 1980 Census earnings data; the updated (1996) GPCI 
employee wage index is computed from 1990 Census data. Employees are grouped into the 
following categories, which are shown with their estimated share of the total physician 
payroll: 2 

Occupational category 1992 GPCI share 19% GPCI share 

Clerical workers 60.4% 44.8% 

Registered nurse 20.4% 25.1% 

Licensed practical nurses 6.9% 9.2% 

Health technicians 12.3% 20.9%. 



'The corresponding shares for the 1992 practice expense GPCI, based on 1987 AMA SMS data, are: 39.1%, 27.6%, and 33.3%, 
respectively. Hence, the share of office rent in practice expense has fallen whereas the shares of employee wages, and especially 
supplies, equipment, and other, have risen 



^The 1992 GPCI shares are estimated from the 1980 Census and the 1983 HCFA Physician Practice Cost and Income 
Survey; the 1996 GPCI shares are estimated from the 1990 Census and the 1988 HCFA Physician Practice Cost and Income 
Survey. 



gpci\pracexp\execsum 



ES-2 



The nurse and technician categories have gained substantially at the expense of clerical 
workers in their share of physician office payroll, primarily because the 1996 shares are 
weighted by practice size whereas the 1992 shares were not. 

A wage index is produced for each of the four employee occupational categories, then 
combined into the overall wage index using the shares shown in trie table. We considered 
four options for computing the wage indices from 1990 Census data: 

1. no change in method of computation from the 1992 practice expense GPCI 
employee wage index; 

2. use earnings of non-manufacturing clerical workers rather than clerical 
workers in all industries; 

3. use mean rather than median hourly earnings; and 

4. use county-specific, rather than metropolitan-wide earnings for large 
metropolitan areas. 

Using wages of non-manufacturing rather than all clerical workers results in very minor 
changes in the employee wage index, which we judged to be too insignificant to warrant 
changing the method of computation. Using mean rather than median earnings resulted in 
somewhat larger, but still minor, changes in the index. We prefer median earnings because 
of its lesser sensitivity to the extremes of the earnings distribution, or to anomalous data. 
Although the employee wage index is also not greatly affected by using county-specific data 
within large metropolitan areas, we believe that this is a justifiable change in the method of 
computing the index. Several populous central city localities (e.g., Manhattan, San Francisco) 
benefit from this change. We believe this is appropriate because physicians in fact have to 
pay higher wages to attract employees in these localities than in outlying parts of their 
metropolitan areas. 

The Health Care Financing Administration concurred with our recommendations 
concerning the employee wage index. The index used in the 1996 practice expense GPCI is 
based on the median hourly earnings of clerical workers, registered nurses, licensed practical 
nurses, and health technicians as measured from the 1990 Census. Earnings are measured for 
counties within large metropolitan areas. 



gpci\ pracexp\ execsum 



ES-3 



In addition, the updated employee wage index is weighted by practice expense RVUs 
rather than by population. The reasons for weighting by RVUs are twofold. First, RVUs 
(which are available by county of location of service) provide a better indication of where 
services are provided. The GPCI should be weighted to reflect input prices at these locations. 
Second, weighting by RVUs guarantees that total Medicare Fee Schedule (MFS) payments for 
practice expense will not change if the payment localities are revised (e.g., if a state with 
multiple payment localities becomes a single statewide locality). Appendix 1-2 shows the 
(minor) differences between the 1996 RVU and population weighted employee wage index. 

The differences between the 1996 practice expense GPCI employee wage index and 
the 1992 practice expense GPCI employee wage index are the use of 1990 rather than 1980 
Census data, county-specific earnings in large urban areas rather than metropolitan area 
earnings, weighting by RVUs rather than population, the classification of all workers by place 
of work rather than some by place of residence, 3 and the use of the Office of Management 
and Budget's June 1993 metropolitan area redefinitions. 4 Appendix 1-3 shows the 1996 
employee wage index, the rescaled 1996 employee wage index, and the 1992 employee wage 
index by Medicare payment locality. The rescaled index is multiplied by the factor 1.00125, 
which was derived by HCFA actuaries to ensure that total Medicare Fee Schedule payments 
for practice expense do not change when the GPCI is updated. The rescaled index will be 
proposed for use in the practice expense GPCI beginning in 1996. 

Appendix 1-4 ranks Medicare payment localities in descending order of difference 
between the 1996 rescaled employee wage index and the 1992 employee wage index. The 
changes from updating are significant. Twenty seven of the 216 localities experience more 
than a 5 percentage point increase and 79 experience more than a 5 percentage point decrease 
in their index values. The largest increase is 24 percentage points, for South Central 
Connecticut; the largest decrease is 18 percentage points, for Peoria, Illinois. Many of the 
largest gainers are in New England, and iriny of the largest losers are in the Midwest. It 
should be remembered, however, that the employee wage index accounts for only about 16 



"In the 1992 GPCI, for sample size reasons, place of work classification was only used in large metropolitan areas. In other 
areas, workers were classified by their place of residence. 

4 

Although the GPCI is eventually calculated for Medicare payment localities, the Census Bureau tabulates the underlying 
earnings data for metropolitan areas and the nonmetropolitan portions of states before it is crosswalked to localities. 



gpci \ pracexpX execsum 



ES-4 



percent of the overall Medicare Geographic Adjustment Factor (GAF). Thus, even the largest 
changes in the employee wage index imply less than a 4 percent change in Medicare fees in 
any payment locality. 

ES4.0 OFFICE RENTAL INDEX 

The office rental index of the practice expense GPCI measures the relative price of 
office space in different areas. Because of the unavailability of valid and reliable physician or 
commercial office rental data for each physician payment locality, relative residential 
apartment rents have been used to proxy relative physician office rents in the GPCI. The 
residential rental data employed is the "Fair Market (Apartment) Rents" (FMRs) established 
by the Department of Housing and Urban Development (HUD) for use in the Section 8 rental 
subsidy program. The current (1992) MFS practice expense GPCI uses 1987 FMRs. An 
adjustment for New York City was made because residential rent control may make relative 
residential rents inaccurate measures of relative commercial rents in New York City. The 
New York City FMR was replaced with the FMR of Bergen-Passaic, New Jersey. 

We analyzed the following issues in the course of updating the GPCI office rental 

index: 

1. data sources available to measure rents; 

2. basing the index on the most recent single year of data versus a three year 
average; 

3. county-specific rather than metropolitan-wide rents in large urban areas; and 

4. rental indices for the New York City area. 

Based on our survey of available data on *-ents, we conclude that the HUD FMRs remain the 
best source of data on which to base the GPCI office rental index. We believe that the FMRs 
provide an acceptable measure of relative office rental costs, unlike any other data source. 
Primary data collection of office rents has the potential to provide a more accurate measure, 
but a survey of commercial rents would be quite expensive, and would need to be conducted 
on an ongoing basis. 



gpci\pracexp\execsum 



ES-5 



r 



The updated 1996 GPCI office rental index could be based on the most recent 
available year of FMRs, or on a multi-year average. A three year average, for example, 
smooths out some of the year to year fluctuations in the FMRs. We recommend that the 
most recent year of FMRs be used in the GPCI. These data are HUD's best estimates of 
current rental market conditions. Moreover, Congress has mandated a two-year transition 
period when the GPCI is updated, which will help smooth fluctuations. The updated 1996 
office rental index is based on the final fiscal year 1994 FMRs, which incorporate HUD's 
benchmark revision based on the 1990 Census, and incorporate OMB s June 1993 
metropolitan area definitions. 

We found that rents do vary systematically within large metropolitan areas. Thus, we 
recommend that county-specific rental data, rather than metropolitan area-wide data, be used 
in large urban areas. Because HUD does not publish county rents within metropolitan areas 
(MSAs), we used a special tabulation of rents reported on the 1990 Census to develop relative 
rental factors for counties within large metropolitan areas (defined as Consolidated MSAs, or 
CMS As). 5 This special tabulation was requested from the Census Bureau by HUD, and is 
consistent with the data HUD uses in establishing its FMRs. 

There are two issues in establishing office rental values for New York City: (1) the 
average rental value for the city, and (2) relative rents in the different New York City 
boroughs (counties). The FMRs could be biased down as a measure of relative New York 
City commercial rents because of residential rent control in the city. However, available 
studies find only a small effect of rent control on median apartment rents in New York City, 
indicating that using the actual HUD FMR for New York City in the GPCI is justifiable. The 
special tabulation of Census data, however, indicates considerable variation in rents within 
New York City, in particular, that rents in Manhattan are much higher than in the other 
boroughs. Since Manhattan is a distinct Medicare payment locality, its higher rents should 
be recognized. This can be accomplished by utilizing county-specific rents within the New 
York City CMSA, as we recommend. 

After reviewing our analyses, HCFA decided that the office rental index of the 1996 
practice expense GPCI should be based on the fiscal year 1994 HUD FMRs, adjusted by 



5 The special tabulation is used only to establish relative rental factors for CMSA counties. The fiscal year 1994 
metropolitan average FMR published by HUD for Primary MSAs is maintained in the 1996 GPCI office rental index. 



gpci\ pracexp\ execsum 



ES-6 



county-specific rental factors within large metropolitan areas (CMSAs) derived from HUD's 
special tabulation of 1990 Census rental data. New York City area rental indices are based 
on actual HUD FMRs, adjusted by the county-specific rental factors. Additionally, for the 
reasons given in Section ES3.0, the office rental index is weighted by practice expense RVUs 
rather than by population. Appendix II-2 shows the minor differenc»?s between the 1996 
RVU- and population-weighted indices. 

Appendix II-3 shows the 1996 office rental index, the rescaled 1996 office rental index, 
and the 1992 office rental index by Medicare payment locality. The rescaled index is 
multiplied by the factor 1.00125, which was derived by the HCFA actuaries to ensure that 
total Medicare Fee Schedule payments for practice expense do not change when the GPCI is 
updated. The rescaled index will be proposed for use in the practice expense GPCI 
beginning in 1996. 

Appendix II-4 ranks Medicare payment localities in descending order of difference 
between the 1996 rescaled office rental index and the 1992 office rental index. The changes 
from updating are significant. Twenty of the 216 Medicare payment localities gain more than 
10 percentage points and 55 lose more than 10 percentage points. The largest gainer is 
Hawaii, whose index rises 52 percentage points, or 41 percent. The largest loser is Victoria, 
Texas, which loses 38 percentage points. Nevertheless, it should be remembered that the 
office rental index accounts for only approximately 10 percent of the overall MFS GAF. 
Hence, even the largest changes in the office rental index imply a gain or loss of less than 5 
percent in the total Medicare fee. 

ES5.0 SUPPLIES, EQUIPMENT AND MISCELLANEOUS PRICES 

Medical supplies, medical equipment, and miscellaneous practice costs comprise about 
one-third of the practice expense GPCI. In the current (1992) GPCI, no adjustment is made 
for geographic variation in the prices of these practice inputs. It has been assumed that 
physicians everywhere could purchase supplies, equipment, and miscellaneous inputs in 
national markets at the same price. However, no data has been located to verify this 
assumption. 

We reevaluated geographic variation in the prices of medical supplies, equipment, 
and miscellaneous inputs through the following activities: 



gpci\ pracexp\ execsum 



ES-7 



• review of previous studies; 

• search for information on geographic variation in prices; 

• interviews of manufacturers, distributors, wholesalers, and purchasers of 
medical supplies and equipment; 

• analysis of shipping costs; 

• analysis of HCFA's 1988 Physician Practice Cost and Income Survey; 

• simulations of the effect of hypothetical input price variation on the overall 
GAF; and 

• evaluation of the benefits and costs of primary data collection of prices by 
area. 

We concluded that there is currently no data available that is adequate to determine 
geographic variation in medical supply, equipment, and miscellaneous input prices. Analysis 
of physician practice expense data and interviews with manufacturers, distributors, and 
purchasers do not clearly establish a geographic pattern of variation in prices. Generally, it 
appears that manufacturers charge uniform prices, aside from shipping costs. Distributors' 
(wholesalers') prices may vary because of local costs of doing business and competitive 
conditions. Larger physician groups are likely to be able to buy through national distributors 
with uniform prices, again aside from shipping costs. Solo practitioners and small group 
physicians are more likely to buy from local distributors with varying prices. Larger groups 
are also likely to receive volume discounts. Shipping costs to Alaska, Hawaii, and Puerto 
Rico appear to be higher than to mainland United States destinations, but no data are 
available on specific differentials between locations of manufacturers/ suppliers/ distributors 
and physician practices in outlying areas. Shipping costs to rural areas do not appear to be 
higher than to nearby urban areas. 

Simulations show that because of the small share of supplies, equipment, and 
miscellaneous in the overall GAF (about 14 percent), moderate price variation in these inputs 
has little effect on the GAF. If the prices of all three inputs varied by as much as 20 percent 
below the national average to 20 percent above the national average across Medicare 
payment localities, the GAF would only change by plus or minus 2.8 percent. Moreover, 
large price differences across areas seem unlikely because supplies and equipment are 



gpci\pracexp\execsum 



ES-8 



physical products that can be transshipped in response to regional price differentials. The 
miscellaneous category is composed of many inputs whose geographic price differences may 
not be highly correlated, limiting the price variation of the overall miscellaneous category. 
To develop a measure of geographic differences in supplies, equipment, and miscellaneous 
prices, an expensive primary data collection effort would be necessary. Improvements in the 
GPCI resulting from primary data collection are likely to be small, bom because of the small 
share of these inputs in the GAF, and because their price variation is likely to be limited. 

Given these considerations, HCFA decided to maintain a policy of no geographic 
adjustment for supplies, equipment, and miscellaneous prices in the GPCI. Thus, this portion 
of the practice expense GPCI is simply 1.000 (i.e., the national average) everywhere. 

ES6.0 THE UPDATED PRACTICE EXPENSE GPCI 

We now discuss the overall 1996 practice expense GPCI, which is the weighted sum 
of its three component indices—employee wages, office rents, and supplies, equipment, and 
miscellaneous prices 6 — as given by the formula in Section ES2.0. The 1996 practice expense 
GPCI is weighted by practice expense RVUs (rather than population), reflects OMB's June 
1993 metropolitan area redefinitions, and captures county-specific price variation within large 
metropolitan areas (CMS As). These are all features— in addition to its use of more recent 
input price data-that distinguish the 1996 practice expense GPCI from the 1992 practice 
expense GPCI. Of these changes, the use of more recent input price information has by far 
the largest overall impact on the index values of Medicare payment localities. Appendix IV-1 
shows that weighting the practice expense GPCI by RVUs versus population has only a 
minor effect on the index values of Medicare payment localities. 

Appendix IV-2 displays the 1996 practice expense GPCI, the 1996 rescaled practice 
expense GPCI, the 1995 rescaled practice expense GPCI, and the 1992 practice expense GPCI 
by Medicare payment locality. HCFA actuaries developed a multiplicative factor (1.00125) 
that was used to ensure that total Medicare Fee Schedule payments for practice expense are 
not affected by the updated practice expense GPCI. The 1996 rescaled GPCI reflects this 



supplies, equipment, and miscellaneous price index is 1.0 everywhe 
gpci\ pracexp\ execsum ES-9 



factor. This rescaled GPCI will be used for payment beginning in 1996. The 1995 rescaled 
GPCI is the simple average of the 1996 rescaled GPCI and the 1992 GPCI. The 1995 rescaled 
GPCI will be used for payment in 1995 only. 

Appendix IV-3 ranks the Medicare payment localities in order of descending 
difference between the 1996 rescaled practice expense GPCI and the 1992 practice expense 
GPCI. The changes in the practice expense GPCI index values from updaLng (1996 versus 
1992 GPCI) are moderate. The largest increase is 12 percent, in Southwest Connecticut. Only 
three payment localites gain more than 10 percent. The largest loss is also 12 percent, in 
Peoria, Illinois. Only three localites lose more than 10 percent. The practice expense GPCI 
comprises 41 percent of the overall Medicare Fee Schedule Geographic Adjustment Factor 
(GAF). Therefore, Medicare fees for a typical service will change by less than 5 percent in 
any area due to updating the practice expense GPCI. Moreover, Congress has mandated a 
two-year transition from the 1992 to the 1996 GPCIs. The change in the 1995 transitional 
GPCI is only half as large as the already small difference between the 1996 and 1992 GPCIs. 



gpci\ pracexp\ execsum 



ES-10 



UPDATING THE GEOGRPAHIC PRACTICE COST INDEX: 
THE PRACTICE EXPENSE GPCI 



1.0 INTRODUCTION 

1.1 Background on the Practice Expense GPCI 

The Geographic Practice Cost Index (GPCI) is used in adjusting Medicare Fee 
Schedule payments for regional variations in physicians' cost of practice. The GPCI is an 
input price index that measures the relative cost of a standard market basket of physician 
practice inputs across areas. The Laspeyres index number form chosen for the GPCI requires 
two basic components: national shares of each practice input in total practice expenses and 
relative price terms for each area. GPCI index values for each area are calculated by 
weighting the relative price terms by the national expense shares to form an average relative 
price for practice inputs in each area. 

Relative prices by area are determined for four practice inputs: physician time, 
nonphysician employee wages, malpractice insurance premiums, and office rent. Geographic 
variation in the price of medical equipment, medical supplies, and miscellaneous is felt to be 
minimal. Congress specified that the GPCI be divided into three components for the 
Medicare Fee Schedule: the physician work GPCI, the practice expense GPCI, and the 
malpractice GPCI. The practice expense GPCI combines the indexes for employee wages and 
office rents, and the cost share for medical supplies, medical equipment, and miscellaneous: 

GPCIPE = 0.391(EMPLOYEE WAGE INDEX) + 0.276(OFFICE RENT INDEX) + .333, 

where the shares of employee compensation, office rent, and supplies, equipment, and 
miscellaneous in practice expenses are 0.391, 0.276, and 0.333, respectively. 1 



These are the current (1992), not updated (1996), physician practice expense shares. 
gpci\ praoexp\ intro I-l 



1.2 Overview of the Report 



This report is divided into four parts, corresponding to the three components of the 
practice expense GPCI plus the overall index: 

Part I: Employee Wage Index; 
Part II: Office Rental Index; 

Part III: Supplies, Equipment, and Miscellaneous Prices; and 
Part IV: Practice Expense GPCI. 

Each part is self-contained, and may be understood separately from the other parts. 
Appendix tables are contained in a companion appendix volume, with parts of the appendix 
corresponding to parts of this report. Alternatives chosen by the Health Care Financing 
Adrrtinistration for the final 1996 payment GPCI are identified and discussed in each part of 
this report. In addition, Part IV assembles the updated (1996) component indices and 
discusses the overall updated (1996) practice expense GPCI. 



gpci\pracexp\intro 



1-2 



PART L 
EMPLOYEE WAGE INDEX 



PART L EMPLOYEE WAGE INDE X 



1.0 INTRODUCTION 

1.1 Background on the Employee Wage Index 

The employee wage index measures the relative cost of compensation of nonphysician 
employees in physicians' offices. Relative cost is measured for each Medicare payment 
locality. Four occupations are represented in the employee index: clerical (administrative 
support), registered nurses, licensed practical nurses, and medical tec hnicians. For the 
current (1992) practice expense GPCI, a wage index for each occupation was developed from 
1980 Census data on the median hourly earnings of these occupations by area. The wage 
indexes were combined according to the estimated share of each occupation in the total 
compensation of nonphysician employee in physicians' offices. These shares were estimated 
using HCFA's 1983 Physician Practice Cost and Income Survey and earnings data from the 
1980 Census. 

1.2 Criticisms of the Employee Wage Index 

The employee wage index has been generally accepted, with only one serious 
criticism directed at it. The major criticism is the age of the 1980 Census data used in the 
1992 GPCI. Some commentators have felt that data of this vintage does not adequately 
represent current wage patterns, especially in areas (e.g., the Northeast) where wages have 
risen rapidly in the 1980s. The 1996 GPCI should answer those concerns, since we update 
the employee wage index with 1990 Census data. 

There are two other possible concerns about the employee wage index. The first is 
with the industry mix of the data used. All-industiy data is used in the current GPCI to 
ensure adequate sample sizes to measure median hourly earnings accurately. That means, 
for example, that clerical workers in manufacturing firms as well as in physicians' offices =ire 
used to measure clerical wages in the GPCI. If the wages of the GPCI occupations differ 
systematically by industry, and industry mix differs by area, the current GPCI could be a 
biased measure of relative wages in physician offices. We conduct a limited test of this 
hypothesis by constructing an employee wage index using wages of clerical workers in 
nonmanuf ac hiring industries, and comparing it to the all-industry GPCI. 



gpci\ pracexp\empl\alltext 



1-1-1 



The second concern is that the use of median earnings for entire Metropolitan 
Statistical Areas (MSAs) may not account for higher wage levels in Medicare localities 
composed of central city areas, e.g., the Manhattan, New York, and Cook County, Illinois 
localities. This concern applies to all the GPCI components based on MSA data (e.g., the 
office rental index and physician work GPCI), not just the employee wage index. In this 
report, we present an employee wage index calculated by county for counties in 
Consolidated Metropolitan Statistical Areas (CMSAs) and compare it to the MSA-based 
index. 

1.3 Overview of Part I of the Report 

The next chapter discusses basic characteristics of the 1990 Census earnings data used 
to update the employee wage index. Sample size distributions by occupation and locality are 
presented. Chapter 3 discusses the methods used to compute the updated employee wage 
indexes, including computation of updated occupational shares for clerical workers, 
registered nurses, licensed practical nurses, and medical technicians. Chapter 4 is the heart 
of the report. Alternative employee wage indexes are presented and evaluated. Differences 
are highlighted by identifying the localities with the largest changes in index values between 
alternatives. Also, a comparison to another major cross-sectional wage index, the PPS 
hospital wage index, is made. 

Chapter 5 discusses the final employee wage index chosen by the Health Care 
Financing Administration based on the earlier analyses contained in this report. This index 
will be used for payment in the 1996 practice expense GPCI. The final (1996) employee wage 
index incorporates the Office of Management and Budget's June 1993 metropolitan area 
redefinitions and is weighted by practice expense relative value units (RVUs) rather than by 
population. 

Appendix II-l shows the alternative employee wage indices discussed in Chapter 4 by 
Medicare payment locality. These indices were constructed before RVU weighting factors 
and OMB's June 1993 metropolitan area redefinitions were available. (They are population 
weighted). Thus, they are not strictly comparable to the final payment indices discussed in 
Chapter 5 or shown in the following appendices. However, the differences between the 
earlier indices and the final 1996 indices are minor and do not affect the conclusions of the 
report. 



gpd\pracexp\empl\alltext 



1-1-2 



Appendix II-2 shows differences by payment locality between the final 1996 employee 
wage index weighted by RVUs and by population. Appendix II-3 displays the employee 
wage indices used in the final (1996) practice expense GPCI. Appendix II-4 ranks payment 
localities by descending order of difference between the 1996 (updated) and 1992 (current) 
employee wage indices. 



gpti\pracexp\ernpl\alltext 



1-1-3 



2.0 DATA FOR THE EMPLOYEE WAGE INDEX 



Potential data sources for the employee wage index were reviewed in Zuckernan, 
Welch, and Pope (1987). It was concluded that decennial Census data were the best available 
for the employee wage index, and 1980 Census data are used in the current (1992) GPCI. 
Since that report, the Physician Payment Review Commission (PPKC, 1991) and the U. S. 
General Accounting Office (GAO, 1993) have reviewed the GPCI and have concluded that the 
Census data is a good choice for use in the employee wage index, although the 1980 Census 
data need to be updated. 

2.1 Description of Census Data 

As part of the decennial Census, questions on income, hours worked, place of work, 
and occupation are asked of roughly 20 percent of the United States population (the exact 
sampling rate varies by type of area). All Census data are self reported. Earnings refer to 
calendar year 1989. Health Economics Research, Inc., contractor to the Health Care Financing 
Aclministration, obtained a special tabulation of 1990 Census earnings data by occupation 
and area from the U.S. Bureau of the Census. The sample is the experienced civilian labor 
force. Earnings are defined as the sum of wage or salary income and net income from self- 
employment. (Nonlabor income is excluded.) Both mean and median hourly earnings are 
tabula*3d, with workers classified by their place of work. For the 1980 Census, place of work 
was tabulated for only one-half of the full sample. Tn 1990, place of work is assigned for 
most workers, so the place of work sample is nearly as large as the place of residence 
sample. All 1990 Census data presented in this report classify workers by place of work. 
Earnings data were obtained for metropolitan areas, counties, states, and the nonmetropolitan 
portions of states. For certain occupations, earnings data were also obtained cross-classified 
by certain industry groupings, such as non-manufac hiring workers. 

2.2 Sample Sizes of Census Data 

Table 2-1 presents the distribution by Medicare payment locality of sample sizes for 
each occupation group used in the non-physician employee wage index. Census provides 
estimates of the total number of people in each occupation/ area cell. We have divided by 6 
to (conservatively) account for the Census "long form" sampling rate of 15-20 percent on 



gpti\pracexp\empl\alltext 



1-2-1 



co 



CO CD oo o> 
m to o) 

O G> iO_ CN 
O" t-~ 1-" CO* 

CO T— T- 



(N 

m 
< 

H 
Q. 

E 
3» 

Cl 
X 

8 

G 



CNl IO O 

t- (N 00 CD 

CO t- CM (O 

O" V-" 



n m co 

o CI t- N 

co m t- co 



t- cn r- m 

O) CD IO 

CM~ 



m 



o> t- o 

t- CO 

CO T- 



O) CO CO O) 
0> I — CM CO 
CO 



N O (O O 
CO 



0) 

■o 
c 



00 



t— CM 0> 

iri oS o 



CM 



CM 



o 

CL 

E 
a> 
c 
55 
o 

s. 

CL 



o 
Cl 
Cl 
=J 

CO 

CD 

g£ 
O CD 

= 1 

o < 



to 




CO 



CD 



« 8 8 

3 o o 
Z £ 'E 
■O OL -g 

£ ? o 

CD CD 

« Si 

D) CD CO 
CD O CD 

Lt □ X 



o 
a, 
O 

I 

CO 

3 

o 
2 
a. 

o 

2 
n 

CO 



1-2-2 



earnings questions. The expenditure share weight of each occupation in. the employee wage 
index is also indicated. 

Sample sizes for administrative support personnel are much greater than for the three 
patient care occupations. The smallest sample size for any locality is 747, and three-fourths 
of all localities had more than 5,000 administrative workers sampled . Random error arising 
from sampling error is not a problem for administrative support workers. 

The sample sizes for registered nurses and health technicians are much smaller. 
However, the smallest locality samples of 40 and 30, respectively, are not so small that we 
would expect median earnings to be significantly affected by the small samples. The sample 
of licensed practical nurses is quite small in some localities, with an estimated minimum 
sample of 3. However, the small sample is a problem for only a few localities; 95 percent of 
the localities have samples larger than 40 (the fifth percentile values, and ninety percent of 
the localities have samples larger than 67 (the tenth percentile value). In addition, with an 
expenditure share of only 9.2% in the wage index, the degree of randomness in the wage 
data for licensed practical nurses would need to be quite large to show a major effect on the 
employee w^ge index. 

We conclude that the Census sample sizes are sufficient to accurately measure median 
hourly earnings by Medicare payment locality, although earnings estimates for the medical 
occupations, especially LPNs, in the least populous localities (such as some of the Texas 
localities) are subject to nontrivial random sampling error. 



gpci\ pracexp\empl\alltext 



1-2-3 



3.0 



METHODS FOR COMPUTING NON-PHYSICIAN EMPLOYEE WAGE INDEXES 



The nonphysician employee wage proxy has two components: (1) employee 
expenditure shares by occupational category, and (2) geographic wage indexes for four 
occupational categories employed by physicians- Section 3 1 discusses determination of the 
occupation shares and compares the updated shares with those ased in constructing the 
current GPCI. Section 3.2 discusses the methodology used fo r creating the alternative 
employee wage indexes for Medicare payment localities 

3.1 Occupation Shares for Employee Occupations 
3.1.1 Data Source 

The data source for the number of employees by occupation category is the 1988 
Physicians' Practice Costs and Income Survey (PPCIS), sponsored by the Health Care 
Financing Administration. The American Medical Association's Physician Masterfile, a 
comprehensive file of all physicians practicing in the U.S., was used as the sampling frame. A 
nationally-representative sample of physicians, stratified by specialty, census division, and 
urban/ rural location, was randomly selected from the Masterfile. To be eligible for the 
survey, physicians had to meet the following criteria: 

1. currently provide patient care services for at least 20 hours per week; 

2. not currently a resident, clinical fellow or research fellow; 

3. not employed by a faculty practice plan, hospital, clinic or HMO, or by a 
federal government agency in 1988; 

4. if in a multipractice arrangement, greater than 80 percent of income from the 
practice in which the physician was full or part owner, or employed by 
another physician or group of physicians; 

5. have spent at least twenty hours per week in patient care (or services) during 
1988; and 

6. have been in the same private practice for all of 1988. 

The final number of completed cases was 3,505, a response rate of 61%. 

Physicians were asked the number of individuals employed by the practice for each of 
eight occupational groups: social workers or psychologists; certified registered nurse 



gpci\ pracexp\empl\ all text 



1-3-1 



anesthetists (CRNAs); registered nurses; licensed practical nurses; medical aides, X-ray and 
laboratory technicians; physician's assistants or nurse practitioners; other patient care 
employees; and administrative and clerical staff. They were then asked how many 
employees in each of these groups worked at least 20 hours per week. 

3.1.2 Computation 

Although the PPCIS asks for eight occupation groups, several occupations are rare in 
physicians' offices and are unavailable or have limited sample sizes in the Census earnings 
data. Thus, we collapsed the eight PPCIS categories into the four used in constructing the 
current employee wage index: administrative and clerical staff, registered nurses, licensed 
practical nurses, and medical technicians. For construction of the cost shares, registered 
nurses, CRNAs, physicians' assistants and nurse practitioners, and social workers and 
psychologists were all included in the registered nurses category. It is unclear what types of 
workers would be included in the PPCIS "other patient care employees" category. This 
category comprised only 3.8 percent of the workers on the survey. To allocate these workers, 
the three patient care categories (RNs, LPNs, and medical technicians) were inflated 
proportionately to absorb the other patient care workers. 

Employee shares were averaged weighted by practice size, with size defined by 
number of nonphysician employees. Weighting by practice size increases the shares of the 
patient care employees relative to administrative employees because larger practices use 
higher proportions of patient care employees. Finally, the e-nployee shares were multiplied 
by the median national hourly earnings of each of the four occupation groups from the 1990 
Census, and the cost share (percent of total payroll) of each group was calculated. 

3.1.3 Comparison with Current Shares 

Table 3-1 presents both the current shares used in calculating the (1992) non-physician 
employee wage index, and the updated shares calculated from the 1988 PPCIS and the 1990 
Census. The administrative and clerical personnel category has the largest share in both the 
current and updated versions. However, the share fell substantially, from 60.4% to 44.8%. 
The three patient care occupation categories all experienced increases in their weights. The 
share of LPNs rose from 6.9% to 9.2%, as did the share of RNs, from 20.4% to 25.1%. Health 
technicians experienced the largest increase in cost shares, from 12.3% to 20.9%. 



gpd\pracexp\empl\ all text 



1-3-2 



TABLE 3-1 



OCCUPATIONAL EXPENDITURE SHARES FOR THE EMPLOYEE WAGE INDEX 



Occupational Group Current Share Updated Share 

Administrative Support Personnel, 60.4 % 44.8 % 

Including Clerical 

Registered Nurses 20.4 25.1 

Licensed Practical Nurses (LPNs) 6.9 9.2 

Health Technicians, Excluding LPNs 12.3 20.9 



SOURCE: 1984 and 1988 Physicians' Practice Costs and Income Survey. 



1-3-3 



gpci\pracexp\empl\TAB3-1 .XLS 



These differences appear to be due to our use of employee counts weighted by 
practice size, since larger practices use a lower proportion of administrative personnel. 
Weighting by practice size is appropriate because the GPCI adjusts per service payments. 
Because larger practices provide more services, they should be weighted more heavily in 
determining the employee shares. In addition, the wages of medical personnel such as 
registered nurses have risen rapidly in the 1980s compared to other occupational groups, 
which would increase medical occupations' shares, because the shares are weighted by 
earnings. 

3.1.4 Validation Versus AMA SMS Shares 

The American Medical Association's Socioeconomic Monitoring System (SMS) survey 
does not collect as many non-physician employee occupations as the PPCIS. It asks only for 
two broad categories: administrative and clerical personnel, and patient care employees. 
However, SMS data can be used to validate the PPCIS administrative versus medical 
employees share. 

Published data from the AMA (AMA 1989, AMA 1991) provide the number of full 
time equivalent (FTE) nonphysician employees per physician and the number of FTE 
administrative and clerical personnel per physician. Assuming the difference between these 
two reflects the number of patient care employees per physician, we can multiply the 
national earnings per hour by the number of FTE emplovees to calculate shares for 
administrative and patient care employees. In 1988, administrative employees accounted for 
49.4 percent of the cost of nonphysician employees per physician according to the SMS data. 
In 1991, this figure had fallen slightly to 48.7 percent. These values are much closer to the 
1988 PPCIS share of 44.8 percent than the earlier share of 60.4 percent. We consider the 
PPCIS shares to be validated by the AMA data. 

3.2 Non-Physician Employee Wage Indexes 

3.2.1 Create County File of Wages for Occupation Categories 

Census earnings data were obtained for metropolitan areas, counties, states, and the 
nonmetropolitan portions of states. Census uses several metropolitan area definitions. 
Metropolitan Statistical Area (MSA) is the most general term. Consolidated Metropolitan 
Statistical Areas (CMSA) are the largest, most populous MSAs. Primary Metropolitan 



gpci\ pracexp\ empl\ all text 



1-3-4 



Statistical Areas (PMSAs) are smaller metropolitan areas that are the constituents of CMSAs. 
In New England, MSAs are not county-based, so Census also defines New England County 
Metropolitan Areas (NECMAs). We began our analysis by creating two files of median 
earnings by occupation. The county is the unit of observation for both files, but, for most 
counties, earnings data is mapped onto counties from a higher geographic level (e.g., median 
metropolitan area earnings mapped onto all counties comprising the MSA). Counties cannot 
be used as the basis for creating the GPCI everywhere because sample sizes are too small to 
permit accurate estimation of earnings for each county. The two analytic files are: 

1. A county level file with metropolitan area and statewide nonmetropolitan 
Census earnings by occupation. Outside of New England, metropolitan areas 
were defined as either Primary Metropolitan Statistical Areas (PMSAs) within 
Consolidated Metropolitan Areas (CMS A), or Metropolitan Statistical Areas 
(MSAs) outside of CMSAs. In New England, metropolitan area was defined as 
New England County Metropolitan Areas (NECMAs) rather than MSAs. 
Nonmetropolitan areas were defined as all counties in a state not located in an 
MSA or NECMA. For example, every county in Wyoming is included as an 
observation on this file. Counties in the Casper MSA (the only metropolitan 
area in Wyoming) are all assigned the Casper MSA median earnings. All other 
counties in the state are assigned Wyoming nonmetropolitan median earnings. 

2. A county level file with county earnings for counties in CMSAs, metropolitan 
area earnings for counties in MSAs, but not in CMSAs, and statewide 
nonmetropolitan earnings for counties not in MSAs or CMSAs. Thus, this file 
is identical to that described in 1. above, except that PMSA earnings m CMSAs 
are replaced with county earnings. Census defines three CMSAs in New 
England— Boston, Hartford, and Providence— but l \\ey are not count} based. 
Three NECMAs match the three New England CMSAs. For the purposes of 
creating this file, we treated the Boston, Massachusetts, Hartford, Connecticut, 
and Providence, Rhode Island NECMAs as CMSAs. That is, we used county 
earnings for counties within these three NECMAs. 1 

Both files described above contain earnings for each occupation, and for all 
stratifications (i.e., industry, education). Two parallel files were then created for mean rather 
than median earnings. All files also contain county population from the 1990 Census. 



'In their June 1993 revision of metropolitan area definitions, Census/ OMB defined only one CMSA in New England: 
Boston. Thus, for the final payment employee wage index, only the Boston NECMA was considered to be equivalent to a 
CMSA. Several towns in Connecticut, mostly in Fairfield county, are part of the New York City CMSA and also part of the 
New Haveri-Bridgeport^tamford-Danbury-Waterbury, Connecticut NECMA. We considered Fairfield county to be part of the 
NECMA, not the CMSA, for purposes of computing the final payment employee wage index. 



gpd\pracexp\empi\alltext 



1-3-5 



3.2.2 Calculate County Index 



County indices were created using two different county weights: population and 
work relative value units (RVUs). To maintain comparability with the 1992 GPCI, and 
because RVUs by county were unavailable when the work was originally done, alternative 
employee wage indices were explored using population weights. Ail employee wage indices 
presented in Chapter 4 are population weighted. The final payment wage indices (Chapter 5, 
Appendices 1-2, 1-3, and 1-4, and Part IV of the report), however, are weighted by RVUs. The 
choice of population or RVU weights has only a small effect on the index values (Appendix 
1-2). The same methods, described now, were used with either weight. 

The population or RVU-weighted national average of the wages assigned to counties 
was taken for each of four occupation groups—administrative support occupations, registered 
nurses (RNs), licensed practical nurses (LPNs), and medical technicians. Although they are 
included on the file, Puerto Rico, the Virgin Islands, and Guam were excluded from this 
national average, as they were in calculating the current (1980 Census based) non-physician 
employee wage index. 2 A county-level index was computed for each occupation by dividing 
each county's median wage by the population or RVU-weighted national average. The 
county-level non-physician employee index was then calculated by multiplying these 
occupation-specific indexes by the occupation cost shares from Table 3-1 and summing. 

3.2.3 Translate County Index to Medicare Payment Localities 

Finally, the county index was translated to Medicare payment localities. Locality 
indexes are population or RVU-weighted averages of their constituent county indexes. If a 
locality includes only part of a county, that county index is included in the locality average 
weighted by the population or RVUs of the county part. If a locality is entirely within one 
county, it is assigned that county's index. 



2 Whether or not Puerto Rico and other outlying areas are included in the national average does not affect the final payment 
index values because the values are rescaled to maintain budget neutrality (i.e., the same aggregate work payments to 
physicians). 



gpci\pracexp\empl\alltext 



1-3-6 



4.0 



ALTERNATIVE NON-PHYSICIAN EMPLOYEE WAGE INDEXES 



This chapter presents four alternatives for updating the current non-physician 
employee wage index (which is based on 1980 Censxis data) with 1990 Census data. The 
alternatives are: 

(1) an update using 1990 Census data / but with no other changes from the current 
GPCI; 

(2) 1990 Census data with the earnings of administrative support staff in non- 
manufacturing industries replacing earnings of administrative support staff in 
all industries; 

(3) 1990 Census data with county-specific earnings for ounties in Consolidated 
Metropolitan Statistical Areas (CMSAs); and 

(4) 1990 Census data using mean earnings rather than median earnings. 

The next four subsections— Sections 4.1 through 4.4~describe the rationale for each of these 
alternatives, and then identify which Medicare payment localities' employee wage indexes 
change the most as a result of changing the method of computation. Finally, as a means of 
evaluating the validity of the updated wage index, in Section 4.5 we compare it to the PPS 
hospital wage index, another major cross-sectional index of wages. 

4.1 Updating With 1990 Census Data 

Updating with the 1990 Census data is the simplest alternative. It has the advantage 
of being consistent with the method by which the current (1992) GPCI was produced. A 
comparison with the 1980-Census-based current employee wage index isolates the effects of 
relative wage changes from 1979 to 1989 from changes in method of computation. 

Updating with newer data, but using the same methodology, results in substantial 
changes in the non-physician employee wage index. Medicare payment localities 
experiencing more than a 7 percentage point gain or loss from updating to the 1990 Census 
data are shown in Table 4-1. Of the 232 localities 3 , 54, or nearly one-quarter, experience more 



3 The GPCI was originally defined for 232 localities. Several multilocality states have recently established single state-wide 
localities, reducing the current number of localities to 216. For comparisons of 1990 Census-based GPCIs to 1980 Census-based 
GPCIs, we use the original 232 localities. For comparisons among alternative GPCIs based on 1990 Census data, we use the 
current 216 localities. 



gpci\ pracexp\ empl\ all text 



1-4-1 



TABLE 4-1 



MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE CHANGES IN THE EMPLOYEE 
WAGE INDEX AS A RESULT OF UPDATING WITH 1990 CENSUS DATA (a) 





Carrier/ 


Updated 


Current 




Percent 


Locality Name 


Locality 


Index 


Index 


Chance 


Chanae 


Larqest Increases 












SOUTH CENTRAL CONN. 


10230/03 


1.270 


1.031 


0.239 


23.2 % 


SW CONNECTICUT 


10230/02 


1.310 


1.076 


0.234 


21.7 


NW AND N. CENTRAL CONN. 


10230/01 


1.231 


1.016 


0.215 


21.2 


EASTERN CONN. 


10230/04 


1.200 


1.000 


0.200 


20.0 


MASSACHUSETTS URBAN 


00700/01 


1.214 


1.022 


0.192 


18.7 


kmaoo o i mi inD^/ni in a i //^ iti r~r*\ 

MASS.SUBURBS/RURAL(CITIES) 


00700/02 


1 .163 


0.993 


0.1 /U 


1 /.2 


RHODE ISLAND 


00870/01 


1.099 


0.955 


0.144 


15.1 


NEW HAMPSHIRE 


00780/40 


1.027 


0.908 


0.119 


13.1 


MnDTUCDM MHA/ ICDOCV 

INUK 1 MtKN NtvV JtKotY 


UUobU/UI 


\ .Z\ 4 


1 .uyo 


u. i i o 


I u.o 


NYC SUBURBS/LONG 1., NY 


00803/02 


1.288 


1.187 


0.101 


8.5 


VERMONT 


00780/50 


0.966 


0.872 


0.094 


10.8 


SANTA CLARA, CA 


00542/09 


1.316 


1.225 


0.091 


7.4 


MIDDLE NEW JERSEY 


00860/02 


1.092 


1.004 


0.088 


8.7 


SOUTHERN MAINE 


21200/03 


0.978 


0.892 


0.086 


9.7 


MANHATTAN, NY 


00803/01 


1.325 


1.245 


0.080 


6.5 


QUEENS, NY 


14330/04 


1.325 


1.245 


0.080 


6.5 


FORT LAUDERDALE, FL 


00590/03 


1.005 


0.928 


0.077 


8.3 


ATLANTA, GA 


01040/01 


1.072 


0.999 


0.073 


7.3 


RICHMOND+CHARLOTTESVL, VA 


10490/01 


1.023 


0.952 


0.071 


7.5 


Larqest Decreases 












PEORIA, IL 


00621/05 


0.903 


1.085 


-0.182 


-16.8 % 


ALASKA 


01020/01 


1.275 


1.437 


-0.162 


-11.3 


ROCK ISLAND, IL 


00621/04 


0.848 


0.976 


-0.128 


-13.1 


DETROIT, Ml 


00710/01 


1.069 


1.196 


-0.127 


-10.6 


NORMAL, IL 


00621/08 


0.874 


0.995 


-0.121 


-12.2 


MILWAUKEE SUBURBS, Wl (SE) 


00951/46 


0.956 


1.075 


-0.119 


-11.1 


DECATUR, IL 


00621/11 


0.836 


0.950 


-0.115 


-12.0 


WHEELING, WV 


16510/17 


0.809 


0.920 


-0.111 


-12.1 


SOUTHERN, IL 


00621/14 


0.800 


0.909 


-0.109 


-12.0 


NORTHWEST, IL 


00621/01 


0.800 


0.909 


-0.109 


-11.9 


QUINCY, IL 


00621/07 


0.800 


0.909 


-0.109 


-11.9 


SOUTHEAST IL 


00621/13 


0.800 


0.909 


-0.109 


-11.9 


NORTH CENTRAL IOWA 


00640/03 


0.806 


0.914 


-0.108 


-11.9 


KANKAKEE, IL 


00621/06 


0.833 


0.941 


-0.108 


-11.5 



1-4-2 



TABLE 4-1 (continued) 



MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE CHANGES IN THE EMPLOYEE 
WAGE INDEX AS A RESULT OF UPDATING WITH 1990 CENSUS DATA (a) 





Carrier/ 


Locality Name 


Locality 


CLEVELAND, OH 


16360/00 


S.CEN. IA(EXCL DES MOINES) 


00640/04 


TOLEDO (LUCAS/WOOD), OH 


16360/00 


NORTHWEST IOWA 


00640/06 


MILWAUKEE, Wl 


00951/04 


DE KALB, IL 


00621/03 


CHAMPAIGN-URBANA, IL 


00621/10 


SE IOWA (INCL IOWA CITY) 


00640/01 


MINNESOTA STATEWIDE 


00720/00 


MONTANA 


00751/01 


SOUTHWEST IOWA 


00640/07 


BROWNSVILLE, TX 


00900/10 


NORTHEAST IOWA 


00640/02 


MICHIGAN, NOT DETROIT 


00710/02 


REST OF MO 


11260/03 


CINCINNATI, OH 


16360/00 


SOUTHERN VALLEY, WV 


16510/20 


OHIO RIVER VALLEY, WV 


16510/19 


CHICAGO, IL 


00621/16 


WYOMING 


00825/21 


EASTERN VALLEY, WV 


16510/18 



Updated Current Percent 
Index Index Change Change 



u.y4y 


1 .Uoo 


-U. I \J i * 


Q ft 


0.777 


0.878 


-0.101 


-I 1 .D 


0.949 


1.048 


-0.099 


-9.4 


0.790 


0.883 


-0.093 


-10.6 


0.982 


1.074 


-0.092 


-8.6 


0.835 


0.927 


-0.092 


-9.9 


0.829 


0.920 


-0.091 


-9.9 


0.841 


0.932 


-0.091 


-9.8 


0.879 


0.969 


-0.090 


-9.3 


0.833 


0.918 


-0.085 


-9.3 


0.815 


0.900 


-0.085 


-9.4 


0.777 


0.859 


-0.082 


-9.6 


0.831 


0.910 


-0.079 


-8.7 


0.959 


1.038 


-0.079 


-7.6 


0.767 


0.843 


-0.076 


-9.1 


0.949 


1.024 


-0.075 


-7.3 


0.821 


0.893 


-0.072 


-8.1 


0.823 


0.895 


-0.072 


-8.1 


1.083 


1.155 


-0.072 


-6.2 


0.884 


0.956 


-0.072 


-7.5 


0.831 


0.902 


-0.071 


-7.9 



(a) more than a seven percentage point gain or loss 



1-4-3 



than a 7 percentage point change. Nineteen, or 9 percent, gain more than 7 percentage 
points, and 35, or 16 percent, lose more than seven percentage points. 

The eight localities experiencing the largest gains, ranging from 12 to 24 percentage 
points, are all in New England. Sixteen localities, many in the Midwest, experience decreases 
of greater than ten percentage points as a result of updating the wage index. Peoria, Illinois 
loses 18 percentage points, the most of any locality. Other Illinois localities, Alaska, Detroit, 
Michigan, and the Milwaukee suburbs are among the largest losers. The geographic 
concentration of the gainers and losers is unlikely if they were the result of random error in 
the Census wage data. 

The magnitude of the wage index changes is not surprising given the ten year lag in 
updating between the 1980 and 1990 Census data. Changes in wage indices can also arise 
because of changes in locality definitions, although the number of localities that has been 
modified is small. For example, the 1992 GPCI for Ohio is based on fifteen separate 
localities, while the 1996 GPCI is based on one consolidated locality encompassing the entire 
state. 

4.2 Non-Manufacturing Administrative Support Staff Wages 

The hourly earnings of an occupation in an area may be affected by industry mix. 
Areas with concentrations of large, unionized manufacturing firms may have higher 
administrative and clerical support earnings than other areas. However, these wages would 
not necessary reflect the conditions found in physicians' offi es, which are smaller firms and 
not unionized. We can test for the effect of industry mix on clerical wages by limiting our 
Census sample for clerical workers to non-manufacturing industries. The occupation share 
for the administrative support staff (at 45 percent) is larger than the share for any other 
occupation group included in the employse wage mdex. Thus, use of administrative wages 
from non-manufacturing industries, rather than all industries, could potentially have 
noticeable effects on the employee wage index in areas that have large concentrations of 
unionized manufacturing workers. 

Table 4-2 presents the largest changes in th° employee wage index by payment 
locality as a result of using clerical wages from non-manufacturing industries compared to all 
industries. The changes are quite minor. Only 6 localities gain more than one percentage 
point; the largest gain is only 1.2 percentage points. Sixteen localities lose more than one 



gpti\pracexp\empl\alltext 



1-4-4 



TABLE 4-2 



MEDICARE PAYMENT LOCALITIES EXPERIENCING THE LARGEST CHANGES IN THE EMPLOYEE 
WAGE INDEX AS A RESULT OF USING WAGES OF NON-MANUFACTURING INDUSTRY 
CLERICAL WORKERS (a) 



Non- 





Carrier/ 


manufacturing 


Current 




Percent 


Locality Name 


Locality 


Index 


Index 


Chanae 


Chanae 


Larcjesi increases 












MIAMUATTAM MV 

MANnA 1 IAN, NY 


UUOUJ/U 1 


1 

I ,ooo 


1 3?5 


0.012 


0.9 % 


UUCCINO, NT 


I 


I .000 


1 3?5 


0.012 


0.9 


MIAIMI pi 

iviirtivii, ri_ 


UUJ3U/U4 


1 054 


1 043 


0.012 


1.1 


U\s ~ 1 V I LJl V r\ OUDUr\DO 


UUJUU/ \J I 


1 201 


1 .189 


0.01 1 


0.9 


MYP c:i IRI IRRC/I own 1 MY 

INTO OUDUr\DO/LUINO 1 . , INT 


nnsrn/n? 


1 ,£,93 


1 288 


0.01 1 


0.9 


Yl IMA A7 


01 rnn/ns 

U I UOVJ/ uu 


f) 877 


0.866 


0.01 1 


1.2 


L-all|Col UCllCaoco 












PDA7P.DIA TY 
DKAZ.UKIA, 1 A 


nnonn/nQ 
uuyuu/uy 


n Q17 






-2.8 % 


ROCHESTER/SURR. CNTYS, NY 


00801/02 


0.966 


0.990 


-0.024 


-2.4 


PEORIA, IL 


00621/05 


0.882 


0.903 


-0.021 


-2.3 


SANTA CLARA, CA 


00542/09 


1.298 


1.316 


-0.018 


-1.4 


GRAYSON, TX 


00900/16 


0.836 


0.853 


-0.018 


-2.1 


MIDLAND, TX 


00900/23 


0.957 


0.972 


-0.016 


-1.6 


BEAUMONT, TX 


00900/20 


0.886 


0.899 


-0.013 


-1.5 


ORANGE, TX 


00900/25 


0.886 


0.899 


-0.013 


-1.5 


OSHKOSH (E CNTRL), Wl 


00951/60 


0.868 


0.881 


-0.013 


-1.5 


METROPOLITAN IN 


00630/01 


0.931 


0.943 


-0.012 


-1.3 


REST OF IN 


00630/03 


0.856 


0.868 


-0 012 


-1.4 


LAKE CHARLES, LA 


00528/04 


0.914 


0.926 


-0.012 


-1.3 


DECATUR, IL 


00621/11 


0.824 


0.836 


-0.011 


-1.4 


MILWAUKEE SURBURBS (SE), Wl 


00951/46 


0.945 


0.956 


-0.011 


-1.2 


ROCKFORD, IL 


00621/02 


0.97* 


0.986 


-0.011 


-1.2 


URBAN IN 


00630/02 


0.876 


0.886 


-0.011 


-1.2 



(a) more than a one percentage point gain or loss 



1-4-5 



percentage point. Three localities lose more than two percentage points: Brazoria, Texas, 
Rochester, New York, and Peoria, Illinois. Four other localities in Texas are among the 
largest losers. We conclude that there is little reason to use non-manufacturing clerical 
earnings rather than all-industry clerical wages in the employee wage index. 

4.3 County-Specific Wages in CMSAs 

The current employee wage index, as well as the two updated options presented 
above, utilize metropolitan-wide median wages in computing Medicare payment locality 
index values. In smaller metropolitan areas this is probably not a significant shortcoming 
because wages are likely to be reasonably homogeneous across the entire metropolitan area. 
However, significant wage variation may exist within large metropolitan areas. The third 
non-physician employee wage index alternative uses county-specific earnings for counties 
within large metropolitan areas, defined as Consolidated Metropolitan Statistical Areas, or 
CMSAs. 

The most useful comparison for this option is the 1990 update without county-specific 
wages (option 1 for updating). This comparison isolates the effects of using county-specific 
wages. The largest differences by payment locality between the updated employee wage 
index and the updated index with county-specific wages are displayed in Table 4-3. 

The changes with county-specific earnings are not very large. The largest increase is 4 
percent, and the largest loss, with one exception, is 5 percent. Ten localities gain more than 
one percentage point and seventeen localities lose more than one percentage point as a result 
of differentiating wages by county within CMSAs. The changes are in the expected direction: 
localities that comprise core areas of large cities gain, and suburban ring localities lose. The 
Manhattan, New York locality (5.8 percentage points) and the Philadelphia/ Pittsburgh, 
Pennsylvania Medical Schools locality (4.9 percentage points) experience the largest gains 
from use of county data within CMSAs. The other gainers, with the exception of San 
Bernadino, California, are all large cities. By far the largest loss is the 14.1 percentage points 
for the Denton, Texas locality. This locality is part of suburban Dallas, and we believe the 
change reflects actual wage differences, not random measurement error. No other locality 
loses more than five percent. 

Although the employee wage index is not greatly affected by using county-specific 
data within CMSAs, we believe that this is a justifiable change in the method of computing 
it. Several populous central city localities would appropriately benefit from this change. We 



gpci\ pracexp\ empl\ all text 



1-4-6 



TABLE 4-3 



MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE CHANGES IN THE EMPLOYEE 
WAGE INDEX AS A RESULT OF USING COUNTY LEVEL DATA WITHIN CMSAS (a) 



Locality Name 

Largest Increases 

MANHATTAN, NY 

PHILLY/PITT MED SHCLS/HOSPS, PA 
SAN FRANCISCO, CA 
MILWAUKEE, Wl 
DALLAS, TX 

SAN BERNADINO/E.CTRL CNTYS CA 
DETROIT, Ml 
HOUSTON, TX 
FORT WORTH, TX 
CHICAGO, IL 



Largest Decreases 

DENTON, TX 
ROCKFORD, IL 

MILWAUKEE SURBURBS (SE), Wl 
NYC SUBURBS/LONG I., NY 
SAN MATEO, CA 
NORTHEAST RURAL TX 
QUEENS, NY 
RIVERSIDE, CA 
SM CITIES (CITY LIMITS) KY 
MARIN/NAPA/SOLANO, CA 
SOUTHEAST RURAL TX 
SOUTHERN NJ 

POUGHKPSIE/N NYC SUBURBS, NY 
MIDDLE NJ 



Carrier/ 


CMSA/County 


Updated 




Percent 


Locality 


Index 


Index 


Change 


Change 


00803/01 


1.383 


1.325 


0.058 


4.4 % 


00865/01 


1.133 


1.084 


0.049 


4.5 


00542/05 


1.361 


1.328 


0.033 


2.5 


00951/04 


1.007 


0.982 


0.025 


2.6 


00900/11 


1.040 


1.020 


0.020 


2.0 


00542/15 


1.108 


1.090 


0.018 


1.7 


00710/01 


1.083 


1.069 


0.014 


1 .3 


00900/18 


1.040 


1.028 


0.012 


1.2 


00900/28 


0.982 


0.970 


0.012 


1.3 


00621/16 


1.094 


1.083 


0.011 


1.0 


00900/12 


0.879 


1.020 


-0.141 


-13.8 % 


00621/02 


0.939 


0. 66 


-0.047 


-4.8 


00951/46 


0.923 


0.956 


-0.033 


-3.4 


00803/02 


1.256 


1.288 


-0.032 


-2.5 


00542/06 


1.297 


1.328 


-0.032 


-2.4 


00900/02 


0.837 


0.868 


-0.030 


-3.5 


14330/04 


1.3CC 


1.325 


-0.025 


-1.9 


00542/27 


1.069 


1.092 


-0.023 


-2.1 


00660/02 


0.848 


0.869 


-0.021 


-2.5 


00542/03 


1.184 


1.206 


-0.021 


-1.8 


00900/03 


0.846 


0.866 


-0.021 


-2.4 


00860/03 


1.077 


1.092 


-0.015 


-1.4 


00803/03 


1.017 


1.030 


-0.013 


-1.3 


00860/02 


1.080 


1.092 


-0.011 


-1.0 



(a) more than a one percentage point gain or loss 



1-4-7 



believe that the wages physicians have to pay for their non-physician employees in these 
central city areas are in fact higher. If county-specific data in CMSAs is used in other GPCI 
price proxies (e.g., the office rental measure and the physician work measure), then A is 
consistent to use it in computing the employee wage index. 

4.4 Mean Earnings 

Our first three alternatives are based on median wages, as is the current employee 
wage index computed from 1980 Census data. Median earnings have been used rather than 
mean earnings because they are less sensitive to the extremes of the earnings distribution. 
Hence, medians are less likely to be affected by anomalous, "outlier" reported wages than are 
means. Especially with small samples, as occurs in some of the less populous localities, 
median earnings are likely to be more stable and have a lower standard error than mean 
earnings. In addition, medians are a better measure of the earnings of the "typical" employee 
than mean earnings. Mean earnings would reflect, for example, a compressed earnings 
distribution due to unionization, or a skewed distribution at upper earnings level due to a 
few high-earning individuals. In general, unusual extremes of the earnings distribution 
should not be reflected in the GPCI. 

The one advantage of mean earnings is that they can be consistently aggregated to 
different geographic levels. For example, state-wide mean earnings are a weighted average 
of inc 1 Lvidual locality mean earnings. Median earnings does not have this property: the 
state-wide median is not a weighted average of the individual locality medians. Mean 
earnings can therefore be scaled consistently from the county level to the state level. This 
could be an advantage if the geographic areas for physician payment are modified, as 
recommended, for example, by the Physician Payment Review Commission (PPRC, 1992). 

Table 4-4 presents the largest changes in employee wage index values comparing the 
median-based and the mean-based indexes. (Both indexes use county-specific data within 
CMSAs.) The changes from using mean earnings are small. Only 3 localities experience 
gains of five or more percentage points: Wausau, Wisconsin (5.2 percentage points), 
Wheeling, West Virginia (5.1 percentage points), and Rockford, Illinois (5.0 percentage 
points). No localities experience a loss of more than five percentage points and only three 
localities San Francisco, California (4.5 percentage points), Salem and other cities, Oregon (4.0 
percentage points) and Massachusetts suburbs/ rural cities (4.0 percentage points) experience 
a loss of four percentage points or more. The reasons for the differences are difficult to 



gpti\praoexp\empl\alltext 



1-4-8 



I 



TABLE 4-4 



MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE CHANGES IN THE EMPLOYEE 
WAGE INDEX AS A RESULT OF USING MEAN RATHER THAN MEDIAN EARNINGS (a) 





Carrier/ 


Means 


Median 




Percent 


Locality Name 


Locality 


Index 


Index 


Chanae 


Chanae 


Laraesf Increases 












WAUSAU (N CNTRL), Wl 


00951/36 


0.905 


0.852 


0.052 


6.1 % 


WHEELING, WV 


16510/17 


0.862 


0.81 1 


0.051 


6.3 


ROCKFORD, IL 


00621/02 


0.989 


0.939 


0.050 


5.3 


PEORIA, IL 


00621/05 


0.954 


0.905 


0.049 


C A 

5.4 


TEXARKANA, TX 


00900/08 


0.902 


0.853 


0.049 


5.8 


BEAUMONT, TX 


00900/20 


0.950 


0.901 


0.049 


C A 

5.4 


ORANGE, TX 


00900/25 


0.950 


0.901 


0.049 


C A 

5.4 


NORMAL, IL 


00621/08 


0.924 


0.876 


0.049 


5.D 


MONTANA 


00751/01 


0.883 


0.835 


0.048 


5. / 


SOUTHERN VALLEY, WV 


16510/20 


0.869 


0.823 


0.046 


5.D 


YUMA, AZ 


01030/08 


0.912 


0.868 


0.044 


5.1 


OHIO RIVER VALLEY, WV 


16510/19 


0.868 


0.825 


u.U4o 


O.o 


SAN ANGELO, TX 


00900/30 


0.854 


0.812 


0.042 


c o 
D.Z 


GRAYSON, TX 


00900/16 


0.897 


0.855 


0.041 


A Q 

4.8 


EASTERN VALLEY, WV 


16510/18 


0.874 


0.833 


0.041 


A f\ 

4.9 


SM E. CITIES, MO 


11260/02 


0.791 


0.757 


0.034 


A C 

4.5 


REST OF MO 


11260/03 


0.799 


0.769 


0.031 


A f\ 

4.0 


ABILENE, TX 


00900/29 


0.866 


0.836 


0.030 


3.6 


SOUTHEAST AL 


00510/03 


0.858 


0.828 


0.030 


3.6 


SOUTH DAKOTA 


00820/02 


0.797 


0.768 


0.029 


3.8 


TUSCON, AZ 


01030/02 


0.949 


0.920 


0.029 


3.1 


DECATUR, IL 


00621/11 


0.866 


0.838 


0.029 


3.4 


ANAHEIM/SANTA ANA, CA 


02050/26 


1 .215 


1 .1 86 


U.uzy 


9 4 


NORTHWEST AL 


00510/01 


0.883 


0.855 


0.028 


3.3 


BIRMINGHAM, AL 


00510/05 


0.954 


0.928 


0.026 


2.8 


ALEXANDRIA, LA 


00528/07 


0.878 


0.853 


0.026 


3.0 


RENO, ET AL. (CITIES), NV 


01290/02 


1.113 


1.089 


0.024 


2.2 


PUERTO RICO 


00973/20 


0.506 


0.482 


0.024 


5.0 


ROCK ISLAND, IL 


00621/04 


0.873 


0.850 


0.024 


2.8 


BATON ROUGE, LA 


00528/03 


0.929 


0.905 


0.024 


2.6 


OKLAHOMA 


01370/00 


0.889 


0.865 


0.023 


2.7 


REST OF AL 


00510/06 


0.846 


0.823 


0.023 


2.8 


RURAL NW COUNTIES, MO 


00740/06 


0.830 


0.808 


0.023 


2.8 


UTAH 


00910/09 


0.899 


0.877 


0.022 


2.6 


MOBILE, AL 


00510/04 


0.860 


0.839 


0.021 


2.5 


SANTA BARBERA, CA 


02050/16 


1.064 


1.043 


0.021 


2.0 


MERCED/SURR. CNTYS, CA 


00542/10 


1.024 


1 003 


0.021 


2.1 


DE KALB, IL 


00621/03 


0.852 


0.831 


0.021 


2.5 



TABLE 4-4 (continued) 



MEDICARE PAYMENT LOCALITIES EXPERIENCING LARGE CHANGES IN THE EMPLOYEE 
WAGE INDEX AS A RESULT OF USING MEAN RATHER I nAN MEDIAN EARNINGS (a) 





Carrier/ 


Means 


Median 




Percent 


Locality Name 


Locality 


Index 


Index 


Change 


Change 


Largest Decreases 












SAN FRANCISCO, CA 


00542/05 


1.316 


1.361 


-0.045 


-3.3 % 


SALEM, ET AL. (CITIES), OR 


01380/03 


0.889 


0.929 


-0.040 


-4.3 


MASS SUBURBS/RURAL CITIES 


00700/02 


1.109 


1.149 


-0.040 


-3.5 


URBAN MASS 


00700/01 


1.176 


1.214 


-0.038 


-3.2 


OAKLAND/EERKLEY, CA 


00542/07 


1.246 


1.283 


-0.037 


-2.9 


EUGENE, ET AL. (CITIES), OR 


01380/02 


0.911 


0.946 


-0.035 


-3.7 


MADISON (DANE CNTY), Wl 


00951/15 


0.955 


0.989 


-0.034 


-3.5 


SW CT 


10230/02 


1.279 


1.313 


-0.034 


-2.6 


REST OF OREGON 


01380/99 


0.907 


0.940 


-0.034 


-3.6 


VERMONT 


00780/50 


0.935 


0.968 


-0.033 


-3.4 


PORTLAND, ET AL. (CITIES), OR 


01380/01 


1.022 


1.052 


-0.031 


-2.9 


NEW HAMPSHIRE 


00780/40 


0.999 


1.029 


-0.030 


-2.9 


SANTA CLARA, CA 


00542/09 


1.289 


1.319 


-0.029 


-2.2 


RHODE ISLAND 


00870/01 


1.068 


1.096 


-0.028 


-2.6 


TEMPLE, TX 


00900/06 


0.848 


0.874 


-0.027 


-3.0 


SW OR CITIES (CITY LIMITS) 


01380/12 


0.885 


0.909 


-0.024 


-2.6 


MANHATTAN, NY 


00803/01 


1.360 


1.384 


-0.024 


-1.7 


SEATTLE (KING CNTY), WA 


00932/02 


1.080 


1.104 


-0.023 


-2.1 


LAS VEGAS, ET AL. (CITIES), NV 


01290/01 


1.002 


1.025 


-0.023 


-2.2 


S. CNTRL CT 


10230/03 


1.251 


1.273 


-0.022 


-1.8 


MILWAUKEE, Wl 


00951/04 


0.986 


1.007 


-0.022 


-2. 1 


SACRAMENTO/SURR. CNTYS, CA 


00542/04 


1.110 


1.131 


-0.021 


-1.9 



(a) more than a two percentage point gain or loss 

Note: Both GPCIs utilize county-specific data within CMSAs. 



1-4-10 



1 J. 



determine without more information on the earnings distributions in the different areas. Less 
urban localities seem to be over-represented among the gainers, whereas some large cities are 
represented among the losers (e.g., San Francisco, Oakland, Manhattan, Seattle). 
Although the median- and means-based indexes do not differ greatly, we prefer the medians- 
based index because of the lesser sensitivity of median earnings to the extremes of the 
earnings distribution. 

4.5 Comparison to the PPS Hospital Wage Index 

The hospital wage index used to compute PPS payments is based on a HCFA survey 
of hospital wage, salary, and compensation data for all hospitals paid under PPS. For each 
MSA and rural area of a state, the index compares the area's average hourly wage to the 
national average hourly wage of all hospital employees. 

There is no strong conceptual basis for adopting the PPS wage index as the price 
proxy for employee wages. The occupational mix of workers employed by hospitals and by 
physicians' offices varies significantly. Almost one half of employee salaries are paid to 
secretaries and administrative staff in physicians' offices, while only about 12 percent of 
hospital labor costs are for these categories of workers (Zuckerman et al., 1987). In addition, 
the PPS wage index is based on a small number of workers (adding to sampling variability) 
and may reflect specific features of the hospital labor market (for example, a hospital with a 
unionized labor force) that may not be relevant to physicians' practices. Finally, the PPS 
wage index makes no adjustments for the occupation mix of hospital workers. 

Although not an alternative to the GPCI employee wage index for payment purposes, 
the PPS wage index is useful in exploring the validity of the employee wage index, especially 
since it is a cross-sectional wage index based on an independent data source. We translated 
the 1988 PPS wage index to Medicare payment localities using our population-weighted 
county to locality crosswalk. Nonmetropolitan counties that have been reclassified as urban 
for purposes of computing their hospitals' PPS payments were assigned the nonmetropolitan 
wage index appropriate for their state. 

We first computed the correlation between the updated employee wage index and the 
PPS wage index. It is 0.93, indicating a high degree of linear association between the two 
indexes. This supports the validity of both indexes. However, the slope coefficient in a 
regression of the GPCI employee wage index on the hospital wage index was only 0.80, 
indicating that the GPCI index is less variable than the PPS wage index. Concomitantly, the 



gpci\ pracexp\ empl\ all text 



1-4-11 



I 



GPCI index has a smaller range than the PPS wage index. The greater range of the PPS 
index could result from its lack of occupation mix corrections, from greater random error in 
less populous areas, or from differences between relative hospital wages and relative wages 
of the GPCI index occupations across areas. 

Table 4-5 shows localities whose hospital wage index differs by more than 7 
percentage points from the employee wage index updated with 1990 Census data. Despite 
the high correlation between the two indexes, several areas have substantially different index 
values, and a large number have moderately different values. Forty-four localities, roughly 
20 percent of all localities, exhibit a difference of more than 7 percentage points between the 
two indexes. However, only 4 areas differ by more than 15 percentage points, and only 15 
by more than 10 percentage points. 

The PPS index exceeds the GPCI index by the largest amount in Temple, Texas. This 
may be due to the teaching hospital there, which could increase the average hospital wage. 
Otherwise, 10 of the 12 localities with higher PPS wage indexes are in California. Apparently, 
hospital workers are paid relatively better in California than the occupations included in the 
GPCI employee wage index. The updated GPCI index exceeds the PPS Index by more than 7 
percentage points in 14 localities. In only two localities is the difference more than 10 
percentage points: Northern New Jersey (13 percentage points) and Atlanta, Georgia (10 
percentage points). 

We conclude that the GPCI employee wage index and the PPS wage index are highly 
correlated and have similar index values for many areas, which supports the validity of both 
indexes. However, index values for a few areas differ signif :antly, and for a larger number 
of areas differ moderately. The precise reasons for these differences cannot be ascertained 
without further investigation, but differences in sample, time period, data source, mean 
earnings (PPS index) versus median earnings (GPCI index), total compensation (PPS index) 
versus wages (GPCI index), and random sampling error probably contribute to the 
differences. The differences do not necessarily indicate that either index is inaccurate, since 
they are designed to measure different things, namely, relative hospital wages (the PPS wage 
index), and relative wages of employees in physicians' offices (GPCI employee wage index). 



gpti\pracexp\empl\alltext 



1-4-12 



TABLE 4-5 



MEDICARE PAYMENT LOCALITIES WITH LARGE DIFFERENCES BETWEEN THE GPCI EMPLOYEE 
WAGE INDEX AND THE PPS WAGE INDEX (a) 



Locality Name 

Largest Increases 



Carrier/ 
Locality 



1988 PPS 
Waoe Index 



Updated 
GPCI 
Index 



Difference 



Percent 
Difference 



Tl~~ i irt| r~ TV/ 

TEMPLE, TX 


uuyuu/uo 


4 1 A "i 
1 .140 


u.o / Z 


n 071 

U.Z / I 


^1 1 % 
o i . i /o 


MONTEREY/SANTA CLARA, CA 


u(Jd4z71z 


1 .zaz 


■i n.71 

I .U / I 


n ooo 
u.zzz 


on 7 

ZU. I 


OAK IT A 1 A A A 

SANTA CLARA, CA 




4 ARO 

\ .40Z 


1 *31 £ 
I .0 i D 


U. ID/ 


19 7 


MARIN/NAPA/SOLANO, OA 


(JU04Z/Uo 


1 .O / U 


1 onfi 


U. 1 oo 




OAKLAND/BERKLEY, CA 


UUd4z7U / 


1 A A E. 


1 OftO 

I .zoz 


U. I OH 


1 p ft 
I z.o 


\/CMTI IDA OA 

VENTURA, CA 


UzUdU/ 1 / 


•1 OftC. 
1 .ZDO 


I . I Z4 


U. It I 


15 fi 

1 £..\J 


MCDPCn/C! IDD PMTVC OA 

MtKUtU/oUKK.L/IN 1 To, L/A 


nni;/i 0/1 n. 
UU04Z/ I U 


•1 IOC 
I . I ZD 


1 ,UU 1 


U. 1 


1? 5 


PORTLAND, ET AL. (Ul 1 IES), UK 


Ul ooU/Ul 


1 1 70 
1 . 1 I Z 


I .U4 / 


n 1 o^ 


11 Q 


OAM rMETOO/IRJIDCDI A 1 OA 

SAN DIEGO/IMPERIAL, OA 


UzUdU/ZO 


1 1 CH 


1 r>70 
i u / z 


n 1 1 ft 

U. HO 


110 

1 1 . VJ 


N. COASTAL CNTYS, CA 


00542/01 


1.194 


1.078 


0.115 


10.7 


CAM CDAM^IC^A OA 

SAN rRANUISUO, OA 


UUd4z7Ud 


1 /t A 1 


1 "JOB 
I OZO 


U. I I J 


fi 7 


SAN MATEO, CA 


00542/06 


1.443 


1.328 


0.115 


8.7 


ALASKA 


01020/01 


1.390 


1.275 


0.114 


9.0 


ABILENE, TX 


00900/29 


0.933 


0.834 


0.099 


11.9 


i/iMrcm II ADC OA 
KINQaS/ 1 ULAKE, OA 


UUd4z71 3 


1 flQR 

i .uyo 


n QQ7 

u.yy / 




Q Q 


SACRAMENTO/SURR. CNTYS, CA 


00542/04 


1 227 


1.129 


0.098 


8.7 


SALEM, ET AL. (CITIES), OR 


01380/03 


1.020 


0.927 


0.094 


10.1 


BROWNSVILLE, TX 


00900/10 


0.870 


0.777 


0.093 


12.0 


ST JOSEPH, MO 


00740/01 


0.949 


0.858 


0.091 


10.6 


SANTA BARBERA, CA 


02050/16 


1.129 


1.040 


0.089 


8.5 


REST OF OREGON 


01380/99 


1.031 


0.946 


0.085 


9.0 


HAWAII 


01120/01 


1.122 


1.039 


0.083 


8.0 


TYLER, TX 


00900/27 


0.974 


0.895 


0.079 


8.8 


MIDLAND, TX 


00900/23 


1.050 


0.972 


0.078 


8.0 


NORTH IDAHO 


05130/12 


0.906 


0.829 


0.077 


9.3 


SOUTH IDAHO 


05130/11 


927 


0.850 


0.077 


9.0 


SW OR CITIES (CITY LIMITS) 


01380/12 


0.983 


0.907 


0.077 


8 5 


PHOENIX, AZ 


01030/01 


1.055 


0.982 


0.073 


7.4 


BEAUMONT, TX 


00900/20 


0.972 


0.899 


0.073 


8.1 


ORANGE, TX 


00900/25 


0.972 


0.899 


0.073 


8.1 


Largest Decreases 












NORTHERN NJ 


00860/01 


1.082 


1.214 


-0.132 


-10.9 % 


ATLANTA, GA 


01040/01 


0.971 


1.072 


-0.101 


-9.4 



1-4-13 



TABLE 4-5 (continued) 



MEDICARE PAYMENT LOCALITIES WITH LARGE DIFFERENCES BETWEEN THE GPCI EMPLOYEE 
WAGE INDEX AND THE PPS WAGE INDEX (a) 



Updated 

Carrier/ 1988 PPS GPCI Percent 
Locality Name Locality Wage Index Index Difference Difference 



SOUTHWEST CT 


10230/02 


1.216 


1.310 


-0.094 


-7.2 


LAFAYETTE, LA 


00528/06 


0.811 


0.894 


-0.083 


-9.3 


NEW ORLEANS, LA 


00528/01 


0.897 


0.979 


-0.082 


-8.4 


DC + MD/VA SUBURBS 


00580/01 


1.107 


1.189 


-0.082 


-6.9 


SOUTH & E. SHORE MD 


00690/03 


0.941 


1.022 


-0.081 


-7.9 


SOUTHERN MAINE 


21200/03 


0.899 


0.978 


-0.079 


-8.1 


LAKE CHARLES, LA 


00528/04 


0.847 


0.926 


-0.078 


-8.5 


REST OF MISSISSIPPI 


10250/01 


0.711 


0.789 


-0.078 


-9.9 


REST OF LA 


00528/50 


0.773 


0.848 


-0.074 


-8.8 


SOUTHEAST AL 


00510/03 


0.753 


0.826 


-0.074 


-8.9 


DALLAS, TX 


00900/1 1 


0.947 


1.020 


-0.072 


-7.1 


DENTON, TX 


00900/12 


0.947 


1.020 


-0.072 


-7.1 



(a) more than a seven percentage point gain or loss 
Note: MSA Data. 



1-4-14 



5.0 



FINAL EMPLOYEE WAGE INDICES 



After reviewing the earlier analyses in this report, the Health Care Financing 
Administration (HCFA) made the following decisions: 

• the clerical portion of the employee wage index should continue to be based 
on wages of administrative support personnel in all industries, not non- 
manufacturing industries only; 

• county-specific earnings should be used within large metropolitan areas 
(CMSAs) rather than metropolitan area earnings; and 

• median hourly earnings rather than mean hourly earnings should continue to 
be used. 

In addition, HCFA decided that the GPCI should be weighted by relative value units 
or RVUs (practice expense RVUs in the case of the employee wage index) rather than by 
population. Weighting by RVUs has two advantages. First, county RVUs identify the 
counties where Medicare services are provided more accurately than county population. The 
GPCI should be weighted to reflect input prices in the locations where services are 
performed. Second, weighting by RVUs assures that aggregate Medicare Fee Schedule 
payments for practice expense will not change if the payment locality configuration is altered 
(e.g., if a state with multiple localities becomes a single statewide locality). The differences 
between the RVU-weighted employee wage index and the population-weighted employee 
wage index are shown in Appendix 1-2, where payment localities are ranked in descending 
order of the difference in the two indices. 4 The change in the employee wage index from 
RVU-weighting ranges from a 3.6 percent increase in South Dakota to a 5.5 percent decline in 
the "rest of Oregon" locality. 

The 1996 updated employee wage index differs from the 1992 employee wa . index 
in the following respects: 



4 The weight affects the computation of the wage index in two places. The first is in normalizing county hourly earnings by 
national average hourly earnings. For the final employee wage index, the national average is weighted by practice expense 
RVUs rather than by population This has the effect of scaling each payment locality's index up or down by the same 
proportion. The second effect occurs in taking a weighted average of county or county part values to form an inde.: value for a 
payment locality. For the final employee wage index, this average is weighted by practice expense RVUs rather than by 
population This change affects each locality's index according to the difference in distribution among counties/ county parts of 
RVUs versus population, and the difference among counties/ county parts of hourly employee earnings. 



gpci\pracexp\empl\ all text 



1-5-1 



• based on 1990 Census earnings data rather than 1980 Census data; 

• reflects OMB's June 1993 redefinitions of metropolitan areas; 5 

• RVU- weigh ted rather than population-weighted; 

• workers are classified by place of work everywhere rather than place of 
residence outside of large metropolitan areas; 6 and 

• county-specific earnings, rather than metropolitan area earnings, are measured 
within large metropolitan areas (CMS As). 

Of these changes, the use of the more recent 1990 Census earnings data has by far the largest 
overall effect on the employee wage index. 

After the original computation of the 1996 practice expense GPCI, HCFA actuaries 
determined that it would have to be rescaled to achieve "budget neutrality", that is, to ensure 
that aggregate payments to physicians are not affected by updating the GPCI. The practice 
expense GPCI was therefore multiplied by the factor 1.00125 and rounded to three decimal 
places. Since the practice expense GPCI is a linear combination of the employee wage and 
office rental indices (plus a constant representing the share of supplies, equipment, and 
miscellaneous in practice expenses), rescaling the practice expense GPCI in effect rescales the 
employee wage index by the same factor. In Appendix 1-3, we show the 1996 employee 
wage index, the rescaled 1996 employee wage index (the 1996 index multiplied by 1.00125), 
and the 1992 employee wage index, by Medicare payment locality. 

Appendix 1-4 ranks Medicare payment localities by descending order of difference 
between the 1996 rescaled employee wage index and the 1992 employee wage index. The 
changes from updating are significant. Twenty seven of the 216 localities experience more 
than a 5 percentage point increase and 79 experience more than a 5 percentage point decrease 
in their index values. The largest increase is 24 percentage points, for South Central 
Connecticut; the largest decrease is 18 percentage points, for Peoria, Illinois. Many of the 
largest gainers are in New England, and many of the largest losers are in the Midwest. It 
should be remembered, however, that the employee wage index accounts for only about 16 
percent of the overall Medicare Geographic Adjustment Factor (GAF). Thus, even the largest 



5 Although the employee wage index is ultimately calculated for Medicare payment localities, the underlying earnings data is 
tabulated by the Census Bureau for metropolitan areas and the nonmetropolitan portions of states before it is crosswalked to 
localities. 



In computing the 1992 employee wage index with 1980 Census data, for sample size reasons, workers were classified by 
place of residence outside of Consolidated Metropolitan Statistical Areas. This was not necessary with the 1990 Census data. 



gpci\pracexp\empl\alltext 



1-5-2 



changes in the employee wage index imply less than a 4 percent change in Medicare fees in 
any payment locality. 



gpci\pracexp\empl\aUtext 



1-5-3 



REFERENCES 



American Medical Association, 1989, Physician Marketplace Statistics 1989, Chicago, IL: 
Center for Health Policy Research. 

American Medical Association, 1991, Physician Marketplace Statistics 1989, Chicago, IL: 
Center for Health Policy Research. 

Dayhoff, D., J.E. Schneider, and G.C. Pope, 1994, Updating the Geographic Practice Cost 
Index: Revised Cost Shares . Final Report to the Health Care Financing Administration 
under Contract No. 500-89-0050. 

Physician Payment Review Commission, 1991, Annual Report to Congress . Washington DC: 
Physician Payment Review Commission. 

Physician Payment Review Commission, 1992, Annual Report to Congress . Washington, DC: 
Physician Payment Review Commission. 

U.S. General Accounting Office, 1993, Medicare Physician Payment: Geographic Adjusters 
Appropriate But Could be Improved with New Data . GAO Report GAO/HRD-93-93. 

Zuckerman, S., W.P. Welch, and G. Pope, 1987, The Development of an Interim Geographic 
Medicare Economic Index . Report to the Health Care Financing Administration under Grants 
No. 18-C-98326/1-01 and 17-C-98758/1-03. 



gpci\pracexp\empl\refs 



I-R-l 



PART a 
OFFICE RENTAL INDEX 



PART IL OFFICE RENTAL INDEX 



1.0 INTRODUCTION 

1.1 Background on the Office Rental Index 

In Medicare's Geographic Practice Cost Index (GPCI), relative prices by area are 
determined for four practice inputs: physician time, nonphysician employee wages, 
malpractice insurance premiums, and office space. The index of office rental prices is a 
component of the practice expense GPCI. Because of the unavailability of valid and reliable 
physician or commercial office rental data for each physician payment locality, relative 
residential apartment rents have been used to proxy relative physician office rents (W elch, 
Zuckerman, and Pope, 1989). Specifically, the residential rental series employed is the "fair 
market rent" (FMR) for a four bedroom apartment established by the Department of Housing 
and Urban Development (HUD). The FMRs are used in the Section 8 rental subsidy 
program. 1987 FMRs were used to establish the rental index used in the current (1992) 
practice expense GPCI. An adjustment for the New York City area was made because 
residential rent control may make relative residential rents inaccurate measures of relative 
commercial rents in New York City. The New York City FMR was replaced with the FMR of 
the Bergen-Passaic, New Jersey area. This was the highest FMR within the New York- 
Northern New Jersey-Long Island consolidated metropolitan area. 

1.2 Criticisms of the Rental Index 

The rental index has generally been regarded as accurate. Nevertheless, certain 
shortcomings or limitations of the FMR proxy have been recognized: 

1. It is based on residential rents, not physician office or commercial rents (PPRC, 
1992). 

2. The New York City FMR may be an inaccurate proxy for relative commercial 
rents because of residential rent control in New York City. Using northern 
New Jersey (Bergen-Passaic) as a proxy for New York City is fairly arbitrary. 
Physician office rents in Manhattan may be higher than in northern New 
Jersey (Schmitz, 1991). 



gpci\ pracexp\ rent\chapl 



II-l-l 



3. There is only a single FMR for each metropolitan area. But substantial 
differences in rents may exist within some metropolitan areas (PPRC, 1992). 
For example, Manhattan may have much higher rents than the Bronx (Schmitz, 

1991) . 

4. The 1987 FMRs need to be updated. 

5. The FMRs for a single year may be misleading because a given area may be in 
a recession or boom period. A three-year moving average may be a better 
measure of the long-run, or typical, rents in an area. 

6. Two studies have compared the FMR proxy to office cost per square foot 
reported on the AMA's physician survey (Gillis et al, 1993; Zuckerman et ai, 

1992) . They have found that actual office expenses vary less than 
proportionally with the FMR proxy. This implies that the FMR proxy may be 
too high in large cities, and too low in rural areas. 

We will address each of these criticisms in Part II of this report. 
1.3 Overview of Part II of the Report 

The next chapter reviews data sources that can be considered for use in constructing 
the office rental index. A discussion of the FMRs is included. We conclude that, although 
imperfect, the FMRs remain the best available data for computing the office rental index. 
However, an expensive primary data collection effort arguably could improve on the FMRs; 
we make some comments on collection of physician or commercial office rents. 

Chapter 3 describes four alternatives we computed to update and refine the office 
rental index. They are: 

(1) Single most recent year (1993) of FMRs; 

(2) Three-year average (1991, 1992, and 1993) of FMRs; 

(3) Single most recent year (1993) of FMRs with county-specific rents in large 
metropolitan areas; and 

(4) Two versus four bedroom FMRs. 

The rationale for each of these alternatives is described, along with methods of computation, 
and Medicare payment localities experiencing large changes from the baseline 1987 FMR 
index. 



gpci\ pracexp\ rent\ chapl 



II-1-2 



Chapter 4 focuses on the New York City area. There are two important issues 
concerning the New York City office rental index. First, as described above, the FMR for 
New York City may be an inaccurate indicator of relative physician office rents because of 
residential rent control there. What is an appropriate rental index for New York City as a 
whole? Second, it has been alle 5 ed that rents vary considerably by borough within New 
York City. Is this true, and if so, can intra-New Y^rV City adjustments be made in the office 
rental index? In Chapter 4, we evaluate several alternatives for New York City area rental 
indices. 

Chapter 5 presents a validation of the FMR proxy using the 1988 HCFA Physician 
Practice Costs and Income Survey (PPCIS). The PPCIS collected self-reported physician office 
expense per square foot for a sample of approximately 3,000 doctors. We make some 
descriptive comparisons of relative FMRs and PPCIS cost per square foot by urban-rural, 
region, and metropolitan area. We also use regression analysis to estimate the overall degree 
of correlation and proportionality between the FMRs and PPCIS cost per square foot. We 
conclude that expense per square foot varies less than proportionally with the FMRs, but that 
this may not indicate any fundamental problem with the FMRs as the basis of the office 
rental index. 

Chapter 6 discusses the final office rental index chosen by the Health Care Financing 
Administration based on the earlier analyses contained in this report. This index will be 
used for payment in the 1996 practice expense GPCI. The final (1996) rental index is based 
on the final fiscal year 1994 FMRs published by H T JD in the Federal Register on April 6, 
1994. These FMRs incorporate HUD's benchmark revision using the 1990 Census, and also 
incorporate the Office of Management and Budget's June 1993 metropolitan area 
redefinitions. The final office rental index is weighted by practice expense relative value 
units (RVUs) rather than by population. 

Appendix II-l shows the alternative office rental indices discussed in Chapter 3 by 
Medicare payment locality. 1 These indices were constructed using 1991, 1992, and 1993 FMR 
data, before the 1994 FMRs, and RVU weighting factors, were available. They do not reflect 
OMB's June 1993 metropolitan area redefinitions, and they are population weighted. Thus, 
they are not strictly comparable to the final payment indices discussed in Chapter 6 or shown 



Except for the two versus four bedroom FMRs contrasted in Section 3.3.4. These two indices were calculated with FY 1994 
FMRs published on October 1, 1993. They reflect OMB's June 1993 MSA redefinitions, but are weighted by population They 
are not directly comparable to the final payment index (Chapter 6), nor to the other rental index variants discussed in Chapter 
3, but the differences are not major. 



gpci\ pracexp\ rent\chapl 



II-1-3 



in the following appendices. However, the differences between the 1993 and earlier indices 
and the final indices based on the 1994 FMRs are not large and do not affect the conclusions 
of the analyses in this report. 

Appendix II-2 shows differences between the final 1996 rental index weighted by 
RVUs and by population for each payment locality. Appendix II-3 displays the office rental 
indices used in the final (1996) practice expense GPCI. Appendix II-4 ranks payment 
localities by descending order of difference between the 1996 (updated) and 1992 (current) 
office rental indices. 



gpci\ pracexp\ rent\ chapl 



II-1-4 



2.0 REVIEW OF DATA SOURCES FOR OFFICE RENTAL INDEX 



This chapter surveys data sources that are candidates for use in the office rental index. 
Several previous data searches and evaluations have been conducted. These are reviewed in 
Section 2.1. Section 2.2 then evaluates four private-sector sources of information on 
commercial office rents by city. Sections 2.3, 2.4, and 2.5 describe three government sources 
of commercial rents— IRS business tax returns, General Services Administration rental surveys, 
and U.S. Postal Service rental information. The Department of Housing and Urban 
Development (HUD) Fair Market Rents (FMRs) are discussed in detail in Section 2.6. Finally, 
primary data collection is considered in Section 2.7, and some conclusions are offered in 
Section 2.8. 

2.1 Previous Data Searches 

Zuckerman, Welch, and Pope (1987) identified three sources of rental information: the 
Building Owners and Managers Association (BOMA) Experience Exchange Report, the 
Institute for Real Estate Management (IREM) annual report, and the HUD FMRs. The 
authors concluded that the BOMA and IREM reports did not have the geographic coverage 
or sufficient sample sizes of buildings to be used for the office rental index. The HUD FMRs, 
on the other hand, were recommended for the GPCI index and comprise the current index. 
Each of these three sources is evaluated in greater detail below. 

Dayhoff and Pope (1990) compared the HUD FMRs with commercial rents tabulated 
by BOMA and the Office Network, and with land prices and a construction cost index. They 
found that the HUD apartment rents were fairly highly correlated with these alternative 
measures, but that the degree of variation differed among the measures. Apartment rents are 
more variable than land prices/ construction costs, but less variable than office rents reported 
by BOMA or the Office Network. The alternatives to the FMRs were only available for a 
limited number of areas. Dayhoff and Pope concluded that the HUD FMRs, while imperfect 
measures of physician office rents, were superior to any of the alternatives. 

Zuckerman, Miller, Wade, and Pauly (1992) recently evaluated data sources for the 
office rental index. They concluded that no source of commercial rents had the necessary 
geographic coverage or accuracy required for the GPCI index. Several sources of residential 
rents exist: the HUD FMRs and median gross rents for renter-occupied units as tabulated by 
the Bureau of the Census. However, unlike the FMRs, the latter data does not control for 



gpci\ pracexp\ rent\ chap2 



II-2-1 



apartment size, and is also based on the entire rental distribution, not just rents paid by 
recent movers. Thus, the FMRs are the best source of residential rents available. Overall, 
Zuckerman et al. conclude that the FMRs remain the best data for use in the GPCI index. 

2.2 Review of Private Sources of Commercial Office Rents 

There are several private sources of office rental rates. Typically, these sources of 
information are used by building owners, managers, real estate brokers, and real estate 
agents to indicate current market conditions. The published material usually contains 
vacancy rates and price per square foot of space (residential, office, or industrial). Four 
sources will be described and evaluated in this section: 

• Building Owners and Managers Association (BOM A) 

• Institute of Real Estate Management (IREM) 

• Society of Industrial and Office Realtors (SIOR) 

• The Urban Land Institute (ULI) 

These organizations are well-known in the real estate industry and are the only sources that 
cover a broad range of geographic areas, although none of them reports data for all 
metropolitan areas or any non-metropolitan areas. They are also the only sources that are 
produced regularly (annually). Other reports may be useful only for analysis of specific cities 
or types of buildings. 

2.2.1 Building Owners and Managers Association 

BOMA International is a private association that publishes information of concern to 
the office building industry. Of particular interest is BOMA Experience and Exchange 
Report : Income/ Expense Analysis for Office Buildings . Published annually, the report 
tabulates operating income and expense data gathered from over 4,700 building owners and 
managers belonging to BOMA. National results are presented according to type of building, 
such as age, height, size, agency managed, single /multiple purpose, and primary use (e.g., 
financial, medical). Results are also presented by major U.S. city. However these tabulations 
distinguish only location (downtown versus suburban) and building size within downtown 
areas. A total of 105 U.S. cities are represented. The number of medical office buildings 



gpci\pracexp\rent\chap2 



II-2-2 



reporting for the entire nation was 37 downtown and 101 suburban. The number of 
buildings reporting in each of the cities ranged from 2 to 98. 

BOMA rents are reported as "gross rental income/' which includes income such as 
parking fees, and are in dollars per square foot of rentable area. BOMA tables show the 
number of buildings reporting, the mean and median rent per square foot, and the range 
(low, high). BOMA also reports estimates of the osts of selected components of the 
operation of office buildings, such as cleaning, maintenance, utilities, and administrative. 
These costs are also reported per square foot of space. 

The national mean office rent, including both downtown and suburban, was reported 
to be $16.87 in 1992 (1991 data). The median rent was $13.28. For medical office buildings 
located downtown, the national mean rent was $19.77, and the median rent was $14.40. For 
medical buildings located in suburban areas of major cities, the mean office rent was $15.66 
and the median was $14.92. For BOMA mean rents for selected cities, refer to Table 2-1. 

2.2.2 Institute of Real Estate Management 

The Institute of Real Estate Management (IREM) produces a report similar to BOMA's 
called the Income and Expense Analysis . IREM publishes data for several types of buildings; 
we will discuss only office buildings. IREM annually collects information from 29 downtown 
metropolitan areas and 55 suburban metropolitan areas. Data are also reported for the entire 
U.S. und by building size and age (medical buildings are not broken cut separately). The 
sample frame is all Certified Property Managers (CPMs), which is a certification granted by 
IREM. 

IREM collects information from building managers on rental income per square foot 
and expenses per square foot. In IREM's report, information is reported as a median, based 
on one or more buildings in a metropolitan area. For rental income, respondents are asked 
to report the "sum of all rents which could have been collected if 100 percent of your 
building had been occupied." The overall median office rent reported by IREM in 1992 (1991 
data) was $14.16 for all downtown buildings (n=491) and $12.26 for all suburban buildings 
(n=1066) (Table 2-1). 



gpti\pracexp\rent\chap2 



II-2-3 



o ip o in in t— co 

CO i O) t- CD O) CD O 
CM t- CM t- t- i- CM 



O 00 



^ — . 
TO TJ T3 C" 



o in co o 



o in 



CM <Q C^l CM CN N CM 



I I I I 



in o o o *- 



co 



t N (N p p 
CN CN CN t- CM CM 



o in 

£8 



CD O 



o in f o 



in ^ —' o in cp oo 

CO CM CO CM CM CM 



in in 

38 



CO 
CN 

1 

. > 

; ! 

e 

, 9; 
' 8. 



o 

co 



2 

3 

'3? 



© 

I 



c 

Q 
— - 

3 

(0 



s o co co ^roinocoooooom 
pcbiri't pi^'cp'^cNT^cpooci 

CM CN t— CMx-CMCMCMCOCMt-CM 



CO 



in «r -<«■ m 

2t tis 



O) q 

CM CM 



>- O ID t- IT) N N 



O t- , O^COOt-OCDO 

cp co co cp r--' -sr csi g co 

CN C*S CNCNr-CNCNCNCOCN 



'I 1^- CM CO CO 00 ' 
1 -tf t- CM CO CM t- ' 



OOOfflt<- 

I 1 cd in t-' s cd 



i i i i 



i i i i 



t- 00 CN O .CM , CO 

cm -^r od ai cm 



2| 00 O 00 CD CD 



i i i 



' $ 1 CD 



00 CM r- O O 

ri s oi 6 t- 
£ *~ *" 



! i (O O) i N , O) i CM 

£° s a i 



2 

ft* 
3 

_a 

3 

CO 



s 
o 
m 



CD 


88 


to 
in 




a : : : 


: : : 8 


CO 


oo 


t 


en 

CM 


18 


I s - 


CM 




in 




' o 


CM 


•-J- CO 


01 


o 


O ■ i ■ 


■ i i 

■ i . 


CN 






co 


co 




CM 


CD 




CO . 


1 

■ f- 


CO 


K in 




s 


CD 

T— 




CO 


oi 




in 


CO 


ci 


CD 


CD 


CN 


8 


in 

T — 



t- n in o cm • i • 
CM o> co co 1 1 ■ 



h- co o co co 

2Sc1?a 



' ' h- 'in 1 •■J- (N co tt cn co o 1 io n 
i ■ r- ' ^ ■ r- tj- cn in'^co 



CO 
CD 



cd h- h- t- co en 
^ co cn ^ tn gi s 

t- t- t- CM CN CM CM 



s 



O 

c o 
■5 "5 _ 

S CO CD 
- O ^ 
c c c "5 

w i- (0 (0 (0 O 

co c raSiOOcowcoco 
£ o g > 

5 8r S<<!<<<<< 
<CQOQJJJJJJJ 



CM 
CO 



CD CO 
CO CD 



CO 
O 

z 

Q 
_J 

3 
ill 



m » - 
CD CD 
C 



c 

CD 

O 



2 
1 
cu 



E 
a> 



2 CO 

C Z 



2 



5 « » £ ■£ | » | 



oi E 



_ <D CD CO _c _C 
2 Z Z Z O 0. 0- 



E ,2 o 

(0 LL. «t 

o •= £ 

co JS 

CO CO CO 



a y 



c .c: 

CO </> 
10 JD 



§5<< 



2 



< 

3 



2 

ro 
c 

CO 

-2 



r 

& 
2 
© 
£ 



E 

Q. 
O 



O /-\ 
(0 <D 

a> £ 



i .5> J= 



O <D *? 



c c 1= 

S 
O 

O) 



C T3 
— 3 

i2 P 



■ ^ VI 

CO Z « 

8 2 & 



(O 



CO 
D 



(5 co" 



P <5 « 

la- 

i< « = 

< -g « 

in 2 i_ 

to c o 

w - * 

O z 2 

* _ co 

c c in -=" = 



£ co 



Q. CO OI 

£ ® £ 2 



1 2 

§8 

£ co <; 



<; 

2 ^ 

CD E 

O « 



3 CO CO 

gl _- 

^ 1 -i 

10 Q Q Q 

in co m m 



c 

CO 

CO 

£ -2 

<D O 

or g 
Q Q 
co co 



» x: 

D) in 

E D) 

<0 



IS "« s 

o E 

s a 

T3 <p 



Q □ 
co d 



LU 

cr 



O 



I 



© 

s 



>2-iS'2S--£-©3£- 



> 
2 

(0 

a. 
E 

o 
O 



O 
CO 



£ 01 



C 3 

— m 

g cS 

5 g 

. . o> 
■e »- 

o i 
Q. 4=- 
© 5 

or uj 



« g 

"5 E 

UJ S) 

© CO 

o c 

c ra 

.2 S 



fro 
© 

O CC 
S."S 
c © 
° 3 

co ■;= 

I? 

10 . - 

co 2 

c T_ 

co to 

5 « 

"I 
co 0_ 



CO 5 3 

UJ = 

§ S J 

o 3 1 

co T- £ 



II-2-4 



2.2.3 Society of Industrial and Office Realtors 



The Society of Industrial and Office Realtors (SIOR) publishes an annual report 
entitled Comparative Statistics of Industrial and Office Real Estate Markets . The report is 
compiled for SIOR by Landauer Associates, Inc., an international real estate consulting firm. 
Landauer collects data by surveying 150 real estate agencies nationwide. This is an 
important distinction from the surveys done by BOMA and IREM, which survey building 
managers. Reported figures are based on responding agencies' own "experience and 
transactions" in the given geographic market. A total of 109 metropolitan areas are 
represented. No separate tabulations are shown by building characteristics. 

Rents are reported for Class A or Class B office space and for space within or outside 
of the boundaries of the central business district. The real estate agencies responding to the 
Landauer survey are asked to report a low, high, and an average rent weighted by the 
amount of vacant space available at the rental rate. Also collected are sales prices per square 
foot. The number of respondents and the number of buildings in each area are not reported. 
Table 2-1 shows SIOR mean office rents for selected U.S. cities. 

2.2.4 The Urban Land Institute 

The Urban Land Institute (ULI) each year publishes Market Profiles, which reports 
average office rents and operating expenses for 35 metropolitan areas. The ULI data is 
collected in the same manner as the SIOR data— each market area is described by a different 
reporting agency. In addition to reporting mean lease rates (dollars per square foot) uid 
mean land prices (also per square foot), the ULI report describes each market area's 
development climate and market conditions for residential, retail, office, industrial and hotel 
markets. Table 2-1 shows ULI mean office rents for selected U.S. cities. 

2.2.5 Limitations of Private Commercial Rental Data 

The principal limitations of the sources of commercial rental data for the GPCI office 
rental index are: 



gpd\pracexp\rent\chap2 



II-2-5 



• Discussions with real estate professionals indicate that these sources reflect 
only corporate office space whereas physicians often locate in residential space 
(Appendix II-B). 

• Sample sizes for each city are very small, which make reported rents highly 
sensitive to the particular buildings or agencies that happen to report. In 
addition, it is unlikely that a small number of buildings can adequately 
represent the average rents of an entire area. 

• The sampling frame used for these surveys may not reflect the "average" 
building; that is, buildings managed by those belonging to professional 
associations may have characteristics different from other buildings. 

• It is likely that building managers choosing to respond to the surveys manage 
buildings that have characteristics different from non-responding buildings. 
The result is that the data under-represent non-responding buildings. 

• None of the commercial rental sources collect data for non-metropolitan areas, 
nor do any of them collect data for all metropolitan areas. City-specific rents 
are available mostly for large metropolitan areas. 

• It is not clear whether respondents are reporting "transaction" rent (the rent 
they actually collect) or "asking" rent (the rent they would collect if (a) the 
space were occupied; or (b) they did not discount their asking price). 

Given their ad hoc methods, the lack of consistency in published rents across sources is not 
surprising (Table 2-1). For example, reported downtown Chicago office rents vary from a 
high of $33.60 per square foot (SIOR) to a low of $18.10 (ULI), a 46 percent difference. For 
Philadelphia, ULI reports an average downtown rent of $30, compared to $16.30 reported by 
IREM (however the IREM reports a median and ULI reports a mean) . Comparing BOMA 
and ULI, which both report means, mean office rent in downtown Atlanta is $13.70 according 
to BOMA and $24.50 according to ULI, an 80 percent difference. These differences are 
probably due to small sample sizes, variation in data collection methodology, and the non- 
randomness of the sample of buildings. 

It is important to emphasize how small the samples are for many of the areas. For 
example, for BOMA the number of downtown reporting buildings for Sacramento was only 
four (Table 2-1). For suburban New York City, BOMA reports data based on only eight 
buildings. IREM based suburban estimates for Philadelphia, Sacramento, and San Francisco 
on eight buildings or fewer. Compounding the small sample size problem for the BOMA 
and IREM surveys is that one respondent can report for a number of buildings. In an area 
where a small number of buildings report, the results can be strongly influenced by the 
experience of a single building owner or manager. Although data for SIOR and ULI are not 



gpci\pracexp\rent\chap2 



II-2-6 



collected using the same methods as BOMA and IREM, they are susceptible to similar 
problems— the experiences of one agency determines the average rent. It is highly probable 
that different data collection methods were used for each market area, and the 
documentation does not reveal the different methods used by the individual agencies. 

Because of the seriousness of these limitations, we do not believe commercial office 
rent data is an appropriate basis for the office ren^ 1 mdex, nor is it of any significant use in 
validating the index. Schmitz (1991) uses the SIOR data to examine New York City area 
rents. Schmitz concluded that the GPCI rental index is as much as 40 percent too low for 
New York City, and that Manhattan office rents are higher than in other boroughs. Although 
we agree with Schmitz's finding that rents in Manhattan are higher than in the other New 
York City boroughs (see Chapter 4), we do not find Schmitz's use of the SIOR data to be 
sound. The realtor who reports SIOR rents for New York City does not believe the SIOR 
data is an accurate indicator of physician office rents thore (see Appendix II-B). As is true of 
the other commercial rental sources, SIOR primarily measures rents for corporate office 
space, whereas physicians in the New York City area typically locate in residential space. 
Appendixes II-A and II-B contain further discussion of the New York City area rents 
reported by the commercial rental sources. 

2.3 Internal Revenue Service Tax Returns 

Physician offices report deductible business rental expenses on their tax returns. 
These data could possibly be used to measure physician office rents by area. GAO and 
HCFA have entered into an agreement with the IRS Statistics of Income Division to obtain 
rental expenses reported on sole proprietorship, partnership, and corporation 1991 tax forms 
of medical doctors, osteopaths, dentists, chiropractors, optometrists, and podiatrists. (The 
statutary definitions of Medicare physicians include these additional practitioners.) 

The advantages of the IRS data are that they are actual expenses from physician 
offices, they should be accurate since inaccuracy in filing tax returns is subject to criminal 
penalties, and they should be available at a much lower cost than a new effort to collect 
physician office rents. However, the IRS data suffer from several serious drawbacks. First, 
rental expenses reported for tax purposes includes equipment lease costs, not just office 
space. Second, square feet of office space is not collected on tax returns. Thus, a rental cost 
per square foot cannot be calculated. An alternative standardization for size of practice 
might be number of FTE employees in a practice. Unfortunately, the number of employees 



gpci\pracexp\rent\chap2 



n-2-7 



that is reported with quarterly FICA payments does not distinguish between full- and part- 
time workers. Third, confidentiality of tax returns might restrict small area analysis. The 
sample size available from the IRS data is not clear at this time. Fourth, no information on 
the detailed specialty of doctors of medicine is available. If rental cost per square foot varies 
by specialty (e.g., internist versus surgeon), and specialty mix varies by area, the IRS measure 
could be biased. 

HCFA and the GAO have acquired state level data reporting rental expenses for 
partnerships and corporations. However, with no method of adjusting for practice size, it is 
impossible to compare per unit costs across practices. (No rental data for sole 
proprietorships was provided.) A more extensive data request has been placed with the IRS 
to provide information including number of employees. This is expected to be available in 
October 1994. Further evaluation of the tax return information can take place at that time. 

Z4 General Services Administration Rental Survey 

Every five years, the General Services Administration (GSA) conducts a survey of the 
rents (cost per square foot) paid for buildings it controls across the country. Utilities are 
included in the rents. Approximately 27,000 buildings are currently included in the survey, 
representing both urban and rural areas. Rents are available by city. 

The GSA has provided HCFA with a file containing average rents by city. Average 
rents were calculated by dividing total rent for buildings within the city by total square feet; 
thuj, they can be thought of as average rent per building, weighted by building size. The file 
also contains average rents by county, calculated in an analogous manner. 

This survey has several shortcomings for use in calculating a nationa 1 rental index for 
use in the GPCI. First, some GSA-rented space— foi example, warehouses—may not be 
representative of physician office rents. Since we have no information on individual 
buildings we cannot exclude those with extremely high or extremely low rents that may be 
indicative of buildings of different types. Second, the mix and location of GSA-rented 
buildings may vary substantially from area to area. For example, if buildings in one couruy 
were all prime office space and buildings in another county were all warehouses, an index 
based on the GSA data would be biased. 

A third shortcoming of the GSA data is that unlike the HUD FMRs, GSA rents are not 
available for all counties. In fact, 33 percent of all urban counties and 69 percent of all rural 
counties are not represented in the GSA rental data. Finally, although the hie contains 



gpci\ pracexp\ rent\chap2 



II-2-8 



information on over 27,000 buildings, five urban counties contain one third of all the 
buildings in the sample. As a result of this concentration, sample sizes are quite low in 
many areas. For example, among rural counties that are in the data, the median sample size 
is one building. 

Given these limitations, the GSA data are not a viable option to replace the FMRs for 
the GPCI rental index. However, they may be useful for validation of relative rents in areas 
for which GSA has a large sample of buildings. Health Economics Research, Inc. (HER) is 
currently evaluating the potential use of these data to validate the GPCI rental index. 

Z5 U.S. Post Office Rental Data 

HCFA has acquired a file from the U.S. Postal Service containing cost and square 
footage data on roughly 35,000 buildings nationwide. The majority of these buildings (over 
28,000) are rented by the Post Office; data for these buildings include the date the lease was 
effective, interior square footage of the building, annual rent, and building location (zip 
code). The remainder of the buildings, which are owned by the federal government, cannot 
be used in construction of an office cost index since the purchase price reflects economic 
conditions at the time of purchase, rather than current conditions. 

Several characteristics of the Post Office data make them more promising than the 
GSA data. First, the file contains data on individual buildings, rather than citywide averages. 
Thus, data can be examined for outlier values that may be inappropriate to include in the 
analysis. For example, a small proportion of buildings were coded as containing zero square 
feet of space. These can be removed from the data before calculating an areawide average 
cost per square foot. Second, although the GSA and Post Office data files contain roughly 
the same number of buildings, the post office buildings are not heavily concentrated in a few 
geographic areas. As a result, sample sizes in most areas are larger than those found with 
the GSA data. Third, the data contain the starting date for the lease on each building. Thus, 
buildings with less recent leases, which are less likely to reflect the current relative rents 
across areas, can be excluded from the sample. 

Despite these advantages relative to the GSA data, txie post office data does not seem 
a viable alternative to replace the HUD FMRs in the GPCI rental index. Sample sizes for the 
post office data are much smaller than for the HUD FMRs, and the post office data do not 
include buildings in all counties. The post office data may, however, be valuable for 



gpci\pracexp\rent\chap2 



II-2-9 



validating FMR relative costs for aggregate levels such *s states or MSA-size classes. HER is 
currently pursuing analysis of these data. 

2.6 Department of Housing and Urban Development (HUD) Fair Market Rents (FMRs) 
2.6.1 General Description 

The Department of Housing and Urban Development (HUD) annually publishes "fair 
market rents" (FMRs) to determine eligibility of rental housing units for the Section 8 housing 
program. HUD estimates FMRs for 339 metropolitan areas and 2,416 nonmetropolitan 
counties. When small sample sizes are a concern in rural counties, HUD groups them to 
achieve an aggregate population of 50,000 or more. FMRs are gross rent estimates, including 
all utilities except telephone. FMRs are defined as the 45th percentile of the rental 
distribution of standard quality rental housing units occupied by recent movers, excluding 
public housing and newly built units. 

FMRs are based primarily on decennial Census data, American Housing Survey 
(AHS) data, Consumer Price Index (CPI) data, and Random Digit Dialing (RDD) surveys of 
rents. The data used to calculate a specific FMR depend on whether or not an area is 
covered by a special metropolitan AHS. 

AHS surveys cover 44 of the largest metropolitan areas, which contain about half of 
all rental housing. The surveys are conducted on a four-year cycle, 11 areas each year. 
Outside these areas, FMRs are based on the decennial Census Ox Housing. Until recently, the 
FMRs were updated each year by the Consumer Price Index for rent and utilities. Separate 
CPIs are available for 74 metropolitan areas, and for the four Census regions. HUD is 
currently phasing in use of RDD surveys of rents in metropolitan and nonmetropo 1 ' m parts 
of the 10 HUD regions to replace the CPIs in updating the FMRs. RDD surveys will also be 
used to annually revise the FMRs of approximately 60 FMR areas (Federal Register, April 30, 
1992, p. 18685). 

HUD publishes FMRs for zero (efficiencies) to four bedroom apartments. Because 
there are more two-bedroom rental units than any other size in most housing markets, 
survey samples of two-bedroom units are larger and, therefore, produce more accurate rent 
estimates. HUD publishes FMRs in the Federal Register twice each year. Proposed estimates 
are usually published in mid-April; final estimates must be available for the federal fiscal 
year beginning on October 1. Public comments (usually from local public housing agencies) 
are received on about 100 of the proposed FMRs each year. To be considered for revision, 



gpci\pracexp\rent\chap2 



II-2-10 



comments must include statistically valid rental housing survey data that justifies the 
recommended changes. FMRs are typically revised for about one-third of the areas on which 
comments are received. 



2.6.2 Strengths of the FMRs for the GPCI Rental Index 



The FMRs have been selected to proxy relative physician office rents in the GPCI 
(Zuckerman, Welch, and Pope, 1987). The FMRs have many compelling advantages. First, 
they are available for all areas using a consistent methodology. Second, they are available for 
relatively small areas (metropolitan areas or rural counties) that should capture most inter- 
area variation in rental costs. Third, they are updated annually. Fourth, they are based on 
large sample sizes of rental units from throughout metropolitan areas or counties. Thus, they 
should be more accurate than other sources derived from smaller samples of buildings, they 
should be representative of the entire area for which they are calculated, and they should 
represent a wide range of building types. Fifth, the FMRs control for rental unit size 
(through the number of bedrooms) and also make first-order corrections for building quality 
(substandard buildings and public housing units are excluded from calculation of the FMRs). 
Holding constant building size and quality reduces bias from comparing dissimilar rental 
units across areas. Sixth, the FMRs are based only on recent movers and so reflect current 
market conditions. Seventh, the FMRs are subject to public comment and revision. They are 
gener:ily regarded as accurate indicators of local housing market conditions. Eighth, the 
FMRs include utility costs in addition to space rental. Ninth, the FMRs are readily available, 
allowing the office rental index to be produced relatively easily and inexpensively, greatly 
reducing administrative burden as opposed to, for example, primary data collection. Tenth, 
although the FMRs are based on self-reported data, respondents probably report accurately 
because most people have to pay their rent monthly and so are well aware of the amount. 
No other data source of which we are aware has all these advantages. 



2.6.3 Weaknesses of the FIvlRs for the GPCI 



Despite their impressive advantages, the FMRs do have some shortcomings for the 
office rental index. First, they obviously lack face validity since they measure residential 
apartment rents, not physician office rents. The use of the FMRs in the GPCI depends on the 
assumption that relative apartment rents are accurate indicators of relative physician office 



gpci\ pracexp\ rent\ chap2 



II-2-11 



I 



rents. We were unable to locate any information or research on the general relation between 
residential and commercial rents (see, however, Chapter 5 for a validation of the FMRs 
against self-reported physician office costs per square foot). However, residential and office 
rents should be affected by the same factors— population density and income in particular— 
and so should be highly correlated. Experts at HUD agreed that residential and commercial 
rents should be closely related, especially in the long run. 

Also, it should not be assumed that generic "commercial" rents are necessarily better 
indicators of relative physician office rents than residential rents. The commercial market is 
comprised of many submarkets: prime downtown corporate office space, suburban retail 
space, warehouse space, manufacturing space, etc. In many areas (e.g., New York City— see 
Appendix II-B) physician offices are located in residential areas and buildings, not 
commercial space. 

Nevertheless, it cannot be denied that the residential and commercial rental markets 
are not identical and that some physicians locate in space that is more "commercial" (e.g., 
medical office buildings) than residential. The magnitude of any distortions in the office 
rental index resulting from the use of residential apartment rents is difficult to determine 
(again, see Chapter 5). Primary data collection of physician office rents on a small area basis 
may be necessary to conclusively settle this issue. 

A second limitation of the FMRs is that they are available only for metropolitan areas 
as a whole. There is evidence of intra- MS A rental variation, especially for large, populous 
MSAs (see Chapters 3 and 4). HUD agrees that rents vary within MSAs, but feels that their 
mission and legislative mandate is to establish rents for market areas as a whole. HUD has 
allowed exceptions to the FMRs for certain areas within particular MSAs. Fortunately, data 
are available to adjust the FMRs by county within MSAs. We analyze sub-MSA rents in 
Chapters 3 and 4. 

A third limitation of the FMRs is that they adjust imperfectly for variations in 
apartment quality across areas. As mentioned above, substandard buildings are excluded 
from calculation of the FMRs. However, no adjustments are made for building quality 
ranging from low-standard to luxury. Residential rents are often higher in wealthier 
suburban rings of major metropolitan areas than in poorer central cities, despite the higher 
population density in central cities. This is one reason HUD is reluctant to decompose the 
FMRs into sub-MSA areas. Presumably apartment quality is higher, on average, in wealthier 
MSAs and parts of MSAs. Wealthier areas may also have a generally more desirable living 
environment (e.g., less crime). Physician office rents may also be higher in higher income 



gpci\ pracexp\ rent\ chap2 



II-2-12 



areas. However, to the extent physician office space is more "standard" across areas than 
apartment space, relative apartment rents unadjusted for quality variations may not perfectly 
track relative physician office rents. HUD's use of the 45th percentile of rents should limit 
the effect of quality differences, since this percentile is less sensitive to a luxury "tail" of the 
rental distribution than, say, the mean rent. 

26.4 Effect of Residential Rent Control on the FMRs 

A fourth disadvantage of the FMRs is potentially the most serious, at least for some 
areas. This is residential rent control, which may cause relative residential rents to diverge 
from relative commercial rents. To assess the impact of rent control on the FMRs, we 
consulted HUD's Report to Congress on Rent Control, September, 1991. 

About 10 percent of all rental housing in the United States is covered by rent control. 
However, "the majority of rent control ordinances appear to have small effects on average 
rent levels" (Report to Congress on Rent Control, p. vi). The most stringent rent control 
programs exist in Berkeley and Santa Monica, California, Brookline and Cambridge, 
Massachusetts, and New York City. 

All these areas except New York City represent only a small proportion of their 
MSAs, counties, or Medicare payment localities. Berkeley comprises a small share of the 
Oakland MSA, which is used to establish the FMR for Berkeley's Medicare payment locality. 
(The locality is the same as the MSA.) Berkeley represents only about 5 percent of the 
population of this locality/MSA, and only about 8 percent of its county (Alameda) 
population. Santa Monica represents only 1 percent of the population of Los Angeles county, 
which is used to establish the office rental index for the eight Los Angeles localities. T f rents 
within Los Angeles county were ever differentiated, Santa Monica's rent control might be 
more of a concern. The office rental index for Brookline and Cambridge Massachusetts is 
based on the Boston MSA FMR. Together Brookline and Cambridge comprise only 5 percent 
of the population of the Boston MSA. Brookline represents 9 percent of the population of its 
county, and Cambridge 7 percent—not much of a problem, even if Boston area rents are 
broken down by county. 

Thus, New York City is the only area where residential rent control could significantly 
bias the FMRs utilized for the office rental index. Two rent control programs are in place in 
New York City. Only about 5 percent of rental units are affected by the older rent control 
program, which is estimated to lower rents below market rates by about 30 percent. The 



gpci\ pracexp\ rent\ chap2 



II-2-13 



45th percentile rent, on which the New York FMR is based, does appear to be affected by the 
less stringent "rent stabilization program." This program is estimated to reduce rents below 
market by about 11 percent (Report to Congress on Rent Control, pp. 15-16). 

We conclude that residential rent control affects the FMRs to a modest degree in New 
York City, but nowhere else. Chapter 4 is devoted to evaluating alternative approaches to 
establishing the office rental index for the New York City area. 

2.7 Primary Data Collection 

Collection of actual physician office rents (cost per square foot) could potentially 
improve the accuracy of the office rental index. The Physician Payment Review Commission 
has called for a survey of commercial rents by the Department of Commerce or the 
Department of Housing and Urban Development (PPRC, 1992). In this section, we make 
several observations about primary data collection. 

A primary data collection effort should focus specifically on office rents of physicians, 
not general commercial rents. The commercial rental market has several segments, some of 
which may be less related to physician office rents than residential rents are. For example, 
prime corporate office space utilized by law, advertising, financial firms, etc. should be 
excluded from the survey because physicians rarely locate in such space. To increase sample 
sizes, similar businesses could be grouped with physician offices. These would include 
dentists, optometrists, chiropractors, and osteopaths. 

The type, and expense, of office space also varies among physicians. Physicians who 
perform many tests and procedures in their offices require extra plumbing, electrical, safety, 
etc. features that can substantially raise the cost per square foot. There is evidence that the 
number of tests and procedures performed in an office setting, and thus physicians' office 
cost per square foot, varies systematically across geographic areas (see Chapter 5). The GPCI 
is an input price index that is designed to measure the cost of a fixed basket of practice 
inputs across areas. Thus, reflecting regional site-of-service differences in the GPCI would be 
undesirable. (A service-specific adjustment for site of service can be incorporated into 
Medicare's physician fees, as HCFA has implemented in a limited fashion and the PPRC has 
proposed modifying and expanding.) A survey of physician office rents should collect 
sufficient information to adjust cost per square foot for differences in the expensiveness of 
construction arising from site-of-service differences. 



gpci\pracexp\rent\chap2 



II-2-14 



A large sample size for the survey is essential. Rents can vary greatly within small 
areas and a broad representation of building locations and types is necessary to ensure the 
representativeness of the collected rents. It is likely that rents for tens of thousands of 
buildings would need to be collected on a national basis. IRS tax return records or the AMA 
Masterfile of physicians could be used to determine the sampling frame. It would be much 
less expensive to piggyback on existing data collect-ion efforts—for example, the Census's 
Service Annual Survey or the IRS's Statistics of Income. 

One alternative would be to expand a physician survey like HCFA's Physician 
Practice Costs and Income Survey (PPCIS) or the AMA's Socioeconomic Monitoring System 
(SMS). Cost per square foot of office space has been collected on both these surveys (see 
Chapter 5 below), but not with a large enough sample size for use in the GPCI. Judging 
from our analyses in Chapters 4 and 5, we would estimate that at least 100 completed 
physician responses per locality would be necessary to achieve an acceptably accurate rental 
index. If the 230 Medicare payment localities were collapsed into 100 larger geographic 
units, this would imply a sample size of 10,000. A sample size of 15,000 to 20,000 would be 
desirable to increase the index's accuracy, but would be quite expensive. Moreover, it would 
have to be repeated to stay abreast of changing market conditions. Of course, other 
information, such as on other practice input prices, could also be collected on a physician 
survey to increase its value per dollar spent. 

An alternative would be to repeat a national survey such as the PPCIS, but 
oversample particular areas of concern such as New York City. An office rental index value 
for New York could be derived as the ratio of the New York median to the national median 
physician office expense per square foot. Oversampling in New York might only increase the 
survey sample size by several hundred. Another possibility would be for HCFA to contract 
with the AMA to oversample New York in their annual SMS survey. 

A final alternative would be to mail a brief questionnaire requesting office expenses 
per square foot to all physicians on the AMA Masterfile. Even with the cooperation of the 
AMA, this approach might achieve a very low response rate. Moreover, respondents might 
not be a random sample of all physicians. Finally, if physicians were informed of the 
purpose of the survey, as presumably they would be, they would have an incentive to 
overreport their costs. 



gpci\ pracexp\ rent\ chap2 



11-2-15 



2.8 Conclusions on Data for the Office Rental Index 

In our judgment, the HUD FMRs remain the best available source for constructing the 
office rental index. We recommend that HCFA incorporate the FY 1994 FMRs into the next 
revision of the GPCI because the FY 1994 FMRs will incorporate a major "benchmark" 
revision based on the 1990 Census and will also incorporate OMB's June 1993 metropolitan 
area redefinitions. 

We believe that the IRS, GSA, and Postal Service rental data warrant further analysis. 
At this time, our judgment is that these data will be more useful to validate and refine the 
FMRs, rather than to replace the FMRs. They may be useful in establishing an office rental 
value for New York City. 

If the various attempts to validate the FMRs as the office rental index indicate 
systematic biases, collection of primary data should be considered. We have discussed 
several possibilities in Section 2.7. Any primary data collection effort is likely to be quite 
expensive and may not improve on the FMRs. If new data needs to be obtained, 
piggybacking on current data collection efforts would be desirable, since new surveys would 
be expensive and would have to be repeated periodically. Less expensive pilot surveys could 
be undertaken to indicate the likelihood that a new survey would improve on the FMRs as 
the basis of the office rental index. 



gpci\ pracexp\ rent\ chap2 



II-2-16 



3.0 



ALTERNATIVE OFFICE RENTAL INDICES 



This chapter presents four alternatives for updating the current office rental index, 
which is based on the fiscal year (FY) 1987 HUD Fair Market Rents (FMRs). Given our 
conclusion in Chapter 2 that the FMRs are the best available data for the office rental index, 
all these alternatives are based on the FMRs. The alternatives are: 

(1) Single most recent year (FY 1993) of FMRs; 

(2) Three-year blend (FY 1993, 1992, and 1991) of FMRs; 

(3) Single most recent year (FY 1993) of FMRs with county-specific rents for 
counties in Consolidated Metropolitan Statistical Areas (CMS As); and 

(4) Two-bedroom versus four-bedroom FMRs. 

In Section 3.1, we present the rationale behind each of these alternatives. Section 3.2 
describes our methods for calculating the office rental proxies. Section 3.3 then discusses 
which localities experience the largest changes in index values as a result of updating with 
the various alternatives. The FY 1993 FMRs were the most recent data available in time for 
our original analyses, and are used in options 1, 2, and 3 above. Option 4 was analyzed with 
the first version of the FY 1994 FMRs produced by HUD and published in the October 1, 
1993 Federal Register . The final FY 1994 FMRs published by HUD in the April 6, 1994 
Federal Register were used in the final 1996 practice expense GPCI computations. 

3.1 Rationale for Office Rental Index Alternatives 

3.1.1 Most Recent Single Year 

Updating with the single most recent year of FMRs is the simplest alternative. It has 
the advantage of utilizing only the most recent data, and is HUD's best estimate of current 
market conditions. The most recent FMRs available for the analyses in this chapter were the 
FY 1993 FMRs. 

3.1.2 Three Year Average 

The annual FMRs are subject to abrupt changes as data from new rental surveys or 
other changes are incorporated. A multi-year average of FMRs should smooth out these 



gpci\ pracexp\ rent\ chap3 



II-3-1 



year-to-year variations and result in a more stable office rental index. We evaluate the most 
recent three-year average of FMRs available in time for our analysis: FY 1993, 1992, and 
1991. 

3.1.3 County-Specific Rents in CMSAs 

A limitation of the FMRs is that only MSA-wide FMRs are established by HUD; no 
intra-MSA rental variation is captured. In smaller metropolitan areas this is probably not a 
significant shortcoming because rents are likely to be reasonably homogeneous across the 
entire metropolitan area. However, significant rental variation may exist within larger 
metropolitan areas. The third office rental index alternative establishes county-specific rents 
for counties within the larger metropolitan areas (Consolidated Metropolitan Statistical Areas, 
or CMSAs). Data from a HUD special tabulation of the 1990 Census is used to calculate the 
county specific rents, as described in more detail below. Otherwise, this option relies on the 
FY 1993 FMRs. 

3.1.4 Two Versus Four Bedroom FMRs 

HUD publishes FMRs annually by metropolitan area and nonmetropolitan county for 
the following apartment sizes: bedrooms (efficiencies), 1 bedroom, 2 bedrooms, 3 
bedrooms, and 4 Dedrooms. Prior to the Fiscal Year 1994 FMR. the FMRs for dm-rent 
bedroom jizes were all based on rental data for 2 bedroom apartments, which are the most 
common across the country. Rents for other bedroom sizes were a constant proportion of the 
2 bedroom rents (less than 1.0 for smaller apartments and more than 1.0 for larger 
apartments). Since the rental index is based on the relative FMRs, which bedroom size was 
used made no difference in the index. 

With the FY94 FMRs, however, HUD began basing the FMRs for different bedroom 
sizes on actual rental data from the 1990 Census and other sources. The FMRs for different 
bedroom sizes are no longer proportional, so it makes a difference which bedroom size is 
chosen for the GPCI rental index. The two most logical choices for the GPCI are the 2 or 4 
bedroom FMRs. The current GPCI index is based on the FY 1987 4 bedroom FMRs, but the 
relative rents for this series are really derived from rental data on 2 bedroom apartments. 

The rental index should be based on the 2 bedroom FMRs, for two reasons. First, the 
current GPCI rental index is really based on 2 bedroom rental data, even though it is 



gpri\pracexp\rent\chap3 



II-3-2 



nominally based on 4 bedroom rents. Thus, using 2 bedroom rents is more consistent with 
what is currently being used in the GPCI arid may lead to less change in the index when 
updating. Second, two bedroom apartments are the most common apartment type across the 
country, and have the largest sample sizes for measuring rental rates. In many areas, four 
bedroom apartments are much less common and four bedroom FMRs are based on much 
smaller sample sizes of rents, which may reflect, for example, an unrepresentative geographic 
distribution of rental units. 

3.2 Methods Used to Calculate Alternative Of 'Ice Rental Indices 

All but one of the office rental indices were computed from the "history" file of FMRs 
supplied to us by the Department of Housing and Urban Development. The one exception is 
the FY 1987 FMR index, which is currently being used as the office rental index in the 
Medicare Fee Schedule. It was provided to us by the Urban Institute. Four bedroom FMRs 
are the basis of all the indexes, except the two bedroom index discussed in Section 3.3.4. 

3.2.1 Create County File of FMRs 

First we created a county file of FMRs, then we translated the county values to 
Medicare payment localities. FMRs were available for counties in all areas except New 
England, where MSAs are not county based. New England FMRs for MS As and 
nonmetropolitan county parts were translated to a county basis using a MSA to county 
crosswalk. Where MSA borders crossed counties, county part FMRs were averaged using 
1990 Census town population weights to form a county value. 

For all but two metropolitan counties, HUD computes FMRs only for MSAs, not for 
constituent counties. Congress requires HUD to establish distinct FMRs for Westchester 
county, New York, which is part of the New York City PMSA, and Monroe county, 
Pennsylvania, which is part of the Scranton-Wilkes-Barre MSA. We retained the distinct 
Westchester FMR on our county file because it should accurately represent rents in that 
county, which is not subject to New York City rent control ordinances. The Monroe county 
FMR was replaced with the Scranton-Wilkes Barre FMR. HUD personnel informed us that 
the Scranton-Wilkes Barre FMR was calculated including Monroe county, although Monroe 
county has its own FMR. Thus, assigning Monroe county the FMR of its MSA treats it 
consistently with other metropolitan counties. Also, as a practical matter, the Monroe county 



gpci\ pracexp\ ren t\ chap3 



II-3-3 



FMR was not established until 1992, so we could not calculate the three-year average proxy 
(1991-1993) for it. 1 HUD also computes a distinct FMR for Columbia city Maryland, part of 
Howard county. Originally, this exception was made because Columbia was part of the 
model cities program. We deleted Columbia city from our FMR file. 2 



3.2.2 Replace New York City FMR 



Once the county file was complete, we assigned the Bergen-Passaic, New Jersey FMR 
to all counties in the New York City PMSA, including Westchester county. Westchester 
county's FMR was replaced so that it would have the same FMR as other counties in its 
PMSA, as do all other metropolitan counties. This is consistent with the current (FY 1987 
FMR-based) GPCI, which uses the Bergen-Passaic FMR for Westchester county. (Westchester 
did not receive its own FMR until 1988. ) 3 



3.2.3 Calculate County FMR Index 



A national average of the county FMRs was taken, weighted by 1990 population. 
Although they are included on the file, Puerto Rico, the Virgin Island, and Guam were 
excluded from this national average, as they were in calculating the current (FY 1987) office 
rental GPCI. 4 A county-level office rental index was computed by dividing each county's 
FMR by the population-weighted national average. 5 



For the FY 1994 FMRs, Monroe county Pennsylvania was classified as part of nonmetropolitan Pennsylvania and thus has 
its own FMR computed by HUD. We used this FMR rather than the Scranton-Wilkes Barre replacement. We continued to use 
the distinct Westchester county New York FMR. 

2 

For the FY 1994 FMRs, HUD deleted 22 counties total from seven metropolitan areas defined by OMB in its June 1993 
revision because HUD believed that these counties had substantially lower rents than the rest of their metropolitan areas. We 
used the FMR rents established by HUD for these 22 counties in the final 19% rental index discussed in Chapter 6. See p. 51410 
of the October 1, 1993 Federal Register . 

3 For the final 1996 office rental index (see Chapter 6), we used the actual HUD FMR for the New York City PMSA, rather 
than replacing it by the Bergen-Passaic FMR. See the discussion in Chapter 4. We also used the Westchester county FMR 
established by HUD for the office rental index used in the 1996 practice expense GPCI. 

vVhether or not Puerto Rico and the other outlying areas are included in the national average has no effect on the final 
GPCI values because they are rescaled to ensure budget neutrality. 

The final office rental index used for payment is weighted by practice expense relative value units (RVUs) rather than 
population, as discussed in Chapter 6. 

gpci\pracexp\rent\chap3 II-3-4 



3.2.4 Translate County Index to Medicare Payment Localities 

Finally, the county index was translated to Medicare payment localities using a 
crosswalk. Locality indexes are population-weighted averages of their constituent county 
indexes. 6 If a locality includes only part of a county, that county index is included in the 
locality average weighted by the population of the county part. 6 If a locality is entirely 
within one county, it is assigned that county's index. 

3.2.5 Create Ccmrity-Specific Rents in CMSAs 

FMRs are produced only for MSAs, not individual metropolitan counties. We 
calculated rental index alternatives that account for rental variation by county within CMSAs. 
CMSAs are composed of smaller metropolitan area units called PMSAs, or Primary 
Metropolitan Statistical Areas. FMRs are defined for each PMSA. 

Our strategy was to maintain the HUD FMR for each PMSA on average, but to 
capture differences in relative rents among counties in each PMSA. To measure intra-PMSA 
rental variation, we obtained a HUD special tabulation of 1990 Census data. This file 
includes residential apartment rents by county. To be consistent with the FMRs, we used the 
45th percentile of the rental distribution of two bedroom apartments rented by recent movers. 
These rents were used to define the relative rent of each county within a PMSA. The 
population- weigh ted average of the county rents was set equal to the FMR produced by 
HUD for the PMSA as a whole. This approach was taken to maintain consistency with the 
FMRs outside of CMSAs, and also to allow easy updating of the rental index as HUD 
produces new FMRs for each PMSA annually. Th<=> relative rents by county obtained from 
the 1990 Census can only be updated every ten years, however, with a new decennial 
Census. 

In New England, CMSAs and PMSAs are not county-based; instead, we employ New 
England County Metropolitan Areas (NECMAs). The Boston, Massachusetts, Hartford, 
Connecticut, and Providence, Rhode Island NECMAs are analogous to the corresponding 
CMSAs. 6 For the purposes of this analysis, we considered these three NECMAs to be 
CMSAs. Our approach requires a single FMR for each of these three NECMAs. (Rents 



OMB's June 1993 metropolitan area redefinitions, Boston is the only New England metropolitan area to be considered a 
CMS A. Thus, for the final 1996 office rental index (Chapter 6), we used county-specific rents only in the Boston NECMA, not in 
the Hartford or Providence NECMAs. 



gpci\ pracexp\ rent\ chap3 



II-3-5 



within NECMAs may vary by county because HUD defines FMRs for MS As, and MS As may 
cross county lines in New England. In cases where an MSA crosses a county line, the county 
value is an average of the MSA FMR and the FMR of the nonmetropolitan part of the 
county.) Thus, before calculating the county-specific rents for the NECMAs, we computed a 
population-weighted average of the FMRs of all counties within each NECMA. Then we 
used the Census rents to define county rents in each NECMA relative to its average FMR. 

3.3 Changes in the Office Rental Index From Updating 
3.3.1 Single Most Recent Yean FY 1993 FMRs 

Updating the office rental index with newer data results in substantial changes. 
Medicare payment localities experiencing more than a 5 percentage point gain or loss from 
updating to the FY 1993 FMRs are shown in Table 3-1. Of the 232 localities, 55, or nearly 
one-quarter, experience more than a 5 percentage point change. Twenty nine, or 13 percent, 
gain more than 5 percentage points, and 26, or 11 percent, lose more than 5 percentage 
points. Hawaii's index increases the most, by 24 percent. Four localities gain more than 20 
percent and 11 more than 10 percent. Miami, Dallas, Seattle, and Washington, D.C. are 
among the biggest gainers. 

Six localities lose more than 20 percent and 11 more than 10 percent. Reno, Nevada 
loses 29 percent, the most of any locality. Phoenix, New Orleans, urban Massachusetts, 
Detroit, and Oakland, California are among the largest losers. Changes in the FMRs may 
occur on a small area basis: Arizona and Texas localities are both among the largest gainers 
and the largest losers. 

The magnitude of these changes is perhaps not surprising given the six year lag in 
updating between the FY 1987 FMRs and the FY 1993 FMRs. However, a number of the 
larger losers seem to be connected with the Random Digit Dialing (RDD) surveys instituted 
by HUD for the first time in FY 1993. Regional HUD field offices recommended areas with 
potential FMR problems, and RDD surveys were conducted for many of these areas. Most of 
the areas with the largest declines in FMRs between 1987 and 1993 were subjects of the RDD 
surveys (Federal Register, April 30, 1992, p. 18685). 

OBRA 1990 requires a two year transition if the GPCIs are not updated annually. 
Therefore, the GPCI in the first year of updating will be a blend of the previous GPCI and 
the updated GPCI. If the FY 1993 FMRs were used to update the office rental index, 
implicitly an average of the 1987 and 1993 FMRs will be incorporated into the practice 



gpci\ pracexp\ rent\ chap3 



II-3-6 



1 



TABLE 3-1 



LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE RENTAL INDEX AS A RESULT OF USING 1993 FAIR 
MARKET RENTS (FMRs) a 









1987 


1993 




Percent 


Locality Name 


Carrier 


Locality 


FMR 


FMR 


Chanqe 


Chanqe 


Largest Increases 














HAWAII 


01120 


01 


1.268 


1.573 


0.305 


24.1 % 


MIAMI, FL 


00590 


04 


1.071 


1.256 


0.185 


17.2 


YUMA (CITY), AZ 


01030 


08 


0.814 


0.998 


0.184 


22.6 


FLAGSTAFF (CITY), AZ 


01030 


05 


0.814 


0.986 


0.172 


21.1 


PRESCOTT (CITY), AZ 


01030 


07 


0.814 


0.986 


0.172 


21.1 


VIRGIN ISLANDS 


00973 


50 


1.000 


1.159 


0.159 


15.9 


DALLAS, TX 


00900 


11 


0.863 


0.998 


0.135 


15.7 


DENTON, TX 


00900 


12 


0.863 


0.998 


0.135 


15.7 


DE KALB, IL 


00621 


03 


0.830 


0.962 


0.132 


15.9 


D.C. + MDA/A SUBURBS 


00580 


01 


1.374 


1.495 


0.121 


8.8 


SMALL E. CITIES, MO 


11260 


02 


0.649 


0.768 


0.119 


18.3 


SEATTLE (KING CNTY), WA 


00932 


02 


1.002 


1.120 


0.118 


11.7 


SOUTH + E. SHORE MD 


00690 


03 


1.015 


1.113 


0.098 


9.6 


POUGHKPSIE/N.NYC SUBURBS 


00803 


03 


1.104 


1.193 


0.089 


8.1 


NORTHERN NEW JERSEY 


00860 


01 


1.339 


1.421 


0.082 


6.1 


RHODE ISLAND 


00870 


01 


1.058 


1.132 


0.074 


7.0 


DELAWARE 


00570 


01 


1.051 


1.122 


0.071 


6.7 


FORT WORTH, TX 


00900 


28 


0.863 


0.932 


0.069 


8.0 


PHILLY/PITT MED SCHS/HOSPS 


00865 


01 


0.985 


1 .050 


0.065 


6.6 


W + SE WA (EXCL SEATTLE) 


00932 


01 


0.939 


1.002 


0.063 


6.7 


VERMONT 


00780 


50 


0.965 


1.027 


0.062 


6.4 


HOUSTON, TX 


00900 


18 


0.7P6 


0.857 


0.061 


7.7 


WESTERN MARYLAND 


00690 


02 


1.004 


1.064 


0.060 


6.0 


CLEVELAND, OH 


16360 


03 


0.811 


0.871 


0.060 


7.4 


SOUTHERN MAINE 


21200 


03 


1.080 


1.139 


0.059 


5.5 


MADISON, Wl (DANE COUNTY) 


00951 


15 


0.942 


1.000 


0.058 


6.1 


SM. TOWN/INDUSTRIAL VA 


10490 


03 


0.763 


0.821 


0.058 


7.6 


SW CONNECTICUT 


10230 


02 


1.397 


1.454 


0.057 


4.1 


SANTA BARBARA, CA 


02050 


16 


1.247 


1.301 


0.054 


4.3 


Largest Decreases 














RENO, ET AL (CITIES), NV 


01290 


02 


1.427 


1.015 


-0.412 


-28.9 


PHOENIX, AZ 


01030 


01 


1.134 


0.884 


-0.250 


-22.0 


TUCSON (CITY), AZ 


01030 


02 


1.084 


0.858 


-0.226 


-20.8 


NEW ORLEANS, LA 


00528 


01 


1.011 


0.791 


-0.220 


-21.7 


ABILENE, TX 


00900 


29 


0.849 


0.639 


-0.210 


-24.7 


SAN ANGELO, TX 


00900 


30 


0.855 


0.678 


-0.177 


-20.7 


MASSACHUSETTS URBAN 


00700 


01 


1.444 


1.304 


-0.140 


-9.7 


DETROIT, Ml 


00710 


01 


1.053 


0.926 


-0.127 


12.0 


OAKLAND-BERKELEY, CA 


00542 


07 


1.580 


1.457 


-0.123 


-7.8 



II-3-7 



gpci\pracexp\rent\TAB3-1 .XLS 



2 



TABLE 3-1 

LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE RENTAL INDEX AS A RESULT OF USING 1993 FAIR 
MARKET RENTS (FMRs) a 









1987 


1993 




Percent 


Locality Name 


Carrier 


Locality 


FMR 


FMR 


Chanqe 


Chanqe 


UTAH 


00910 


09 


0.917 


0.798 


-0.119 


-13.0 


OKLAHOMA 


01370 


00 


0.799 


0.684 


-0.115 


-14.4 


SANTA CLARA, CA 


00542 


09 


1.719 


1.611 


-0.108 


-6.3 


TEMPLE, TX 


00900 


06 


0.773 


0.676 


-0.097 


-12.6 


ALASKA 


01020 


01 


1.304 


1.215 


-0.089 


-6.8 


E.CEN+NE WA (EXCL SPOKANE) 


00932 


04 


0.932 


0.847 


-0.085 


-9.1 


SPOKANE+RICHLND(CITIES) 


00932 


03 


0.970 


0.888 


-0.082 


-8.5 


REST OF NEVADA 


01290 


99 


1.221 


1.144 


-0.077 


-6.3 


SOUTHEAST IL 


00621 


13 


0.751 


0.675 


-0.076 


-10.1 


ANAHEIM-SANTA ANA, CA 


02050 


26 


1.618 


1.546 


-0.072 


-4.4 


PUERTO RICO 


00973 


20 


0.787 


0.716 


-0.071 


-9.0 


ATLANTA, GA 


01040 


01 


1.083 


1.017 


-0.066 


-6.1 


PORTLAND, ET AL (CITIES), OR 


01380 


01 


1.002 


0.936 


-0.066 


-6.6 


QUINCY, IL 


00621 


07 


0.751 


0.687 


-0.064 


-8.5 


SOUTHEAST (OHIO VALLEY) OH 


16360 


15 


0.769 


0.714 


-0.055 


-7.1 


WESTERN RURAL TEXAS 


00900 


04 


0.679 


0.625 


-0.054 


-7.9 


SOUTHERN IL 


00621 


14 


0.751 


0.699 


-0.052 


-6.9 



a More than a 5 percentage point gain or loss. 



11-3-8 



gpci\pracexp\rent\TAB3-1 XLS 



expense GPCI in the first year of updating. Changes resulting from this blend are shown in 
Table 3-2. The number of localities with a large gain or loss is cut in half compared with a 
FY 1993 FMR update: 24 localities gain or lose more than 5 percentage points compared with 
55. As expected, the largest gains or losses are also cut in half. Of course, with a two-year 
transition to a fully updated GPCI, these same changes would occur in the second transition 
year as well as the first. 

3.3.2 Three Year Average: FY 1993, 1992, 1991 FMRs 

One way to increase the stability of the office rental index is to employ a multi-year 
average. Table 3-3 shows large changes in locality rental indices as a result of updating with 
an average of FY 1993, 1992, and 1991 FMRs. As expected, large gains or loses are 
attenuated. As compared to a FY 1993 FMR update, Hawaii gains 15 percent instead of 24 
percent, and Reno, Nevada loses 11 percent instead of 29 percent. The number of localities 
experiencing gains of more than 5 percentage points falls only from 29 to 22, and loses of 
more than 5 percentage points only from 26 to 23. Thus, the three year average attenuates 
large gains/ losses more it than reduces the number of localities with moderate to large 
changes. Overall, the total (post-transition) changes from a three-year average update have 
some similarity to the first-year changes resulting from a FY 1993 FMR update. 

3.3.3 County-Specific Rents in CMSAs 

Our third option for updating the office rental index uses the FY 1993 FMRs and 
county-specific rents in CMSAs. The most useful comparison for this option is the FY 1993 
FMR update without county-specific rents (option 1 for updating). Comparison to FY 1993 
FMRs isolates the effects of using county-specific rents from the effects of updating with 
newer data. Differences between the FY 1993 FMR proxy and the 1993 FMRs with county- 
specific rents are displayed by Medicare payment locality in Table 3-4. 

Few localities experience large changes in their office rental index as a result of 
differentiating rents by county within CMSAs. Only 7 rental indices change by more than 5 
percentage points, with 5 gaining and 2 losing. By far the largest change is a 29 percent gain 
for the Manhattan, New York locality. No other locality's index changes by more than 10 
percent. The Queens borough of New York City gains 8 percent, while the locality 



gpd\praoexp\rent\chap3 



II-3-9 



TABLE 3-2 



LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE RENTAL INDEX AS A RESULT OF USING 
A BLEND OF 1987 AND 1993 FAIR MARKET RENTS (FMRs) a 



1987 1987,1993 Percent 



Locality Name 


Carrier 


Locality 


FMR 


FMR 


Chanqe 


Chanqi 


Largest Increases 














HAWAII 


01120 


01 


1.268 


1.421 


0.153 


12.0 


YUMA (CITY), AZ 


01030 


08 


0.814 


0.906 


0.092 


11.3 


MIAMI, FL 


00590 


04 


1.071 


1.163 


0.092 


8.6 


FLAGSTAFF (CITY), AZ 


01030 


05 


0.814 


0.900 


0.086 


10.6 


PRESCOTT (CITY), AZ 


01030 


07 


0.814 


0.900 


0.086 


10.6 


VIRGIN ISLANDS 


00973 


50 


1.000 


1.079 


0.079 


7.9 


DALLAS, TX 


00900 


11 


0.863 


0.931 


0.068 


7.8 


DENTON, TX 


00900 


12 


0.863 


0.931 


0.068 


7.8 


DE KALB, IL 


00621 


03 


0.830 


0.896 


0.066 


8.0 


D.C. + MDA/A SUBURBS 


00580 


01 


1.374 


1.434 


0.060 


4.4 


SEATTLE (KING CNTY), WA 


00932 


02 


1.002 


1.061 


0.059 


5.9 


SMALL E. CITIES, MO 


11260 


02 


0.649 


0.708 


0.059 


9.2 


Largest Decreases 














RENO, ETAL (CITIES), NV 


01290 


02 


1.427 


1.221 


-0.206 


-14.4 


PHOENIX, AZ 


01030 


01 


1.134 


1.009 


-0.125 


-11.0 


TUCSON (CITY), AZ 


01030 


02 


1.084 


0.971 


-0.113 


-10.4 


NEW ORLEANS, LA 


00528 


01 


1.011 


0.901 


-0.110 


-10.9 


ABILENE, TX 


00900 


29 


0.849 


0.744 


-0.105 


-12.4 


SAN ANGELO, TX 


00900 


30 


0.855 


0.767 


-0.088 


-10.3 


MASSACHUSETTS URBAN 


00700 


01 


1.444 


1.374 


-0.G/0 


-4.9 


DETROIT, Ml 


00710 


01 


1.053 


0.990 


-0.063 


-6.0 


OAKLAND-BERKELEY, CA 


00542 


07 


1.580 


1.519 


-0.061 


-3.9 


UTAH 


00910 


09 


0.917 


0.857 


-0.060 


-6.5 


OKLAHOMA 


01370 


00 


0.799 


0.742 


-0.^7 


-7.2 


SANTA CLARA, CA 


00542 


09 


1.719 


1.665 


-0.C + 


-3.1 



a More than a 5 percentage point gain or loss. 



II-3-10 



gpci\pracexp\rent\TAB3-2.XLS 



1 



TABLE 3-3 



LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE RENTAL INDEX AS A RESULT OF USING 
A BLEND OF 1991, 1992, AND 1993 FAIR MARKET RENTS (FMRs) (a) 









4A07 

1987 


•» J .0> 'GO 

91 , 32, 93 




Percent 


Locality Name 


Carrier 


Locality 


FMR 


FMR 


Chanqe 


Chanqe 


Largest Increases 














HAWAII 


ft a a ft ft 

01120 


ft A 

01 


A ft ft O 

1 .268 


A A C C 

1 .455 


ft A 0~7 

0.187 


14.8 % 


\/| IMA //*"\ 1 -r\A A ~7 

YUMA (CITY), AZ 


01030 


ft ft 

08 


0.814 


ft ftft~7 

0.997 


ft A ft ft 

0.183 


22.5 


FLAGSTAFF (CITY), AZ 


01030 


ft f~ 

05 


0.814 


0.982 


0.168 


ft ft ~7 

20.7 


nnro^ATT //mt\a a ~~t 

PRESCOTT (CITY), AZ 


01030 


07 


0.814 


0.982 


0.168 


20.7 


VIRGIN ISLANDS 


00973 


50 


1 .000 


1.152 


0.152 


15.2 


DE KALB, IL 


00621 


03 


0.830 


0.966 


0.1 36 


Aft A 

16.4 


SEATTLE (KING CNTY), WA 


00932 


02 


1 .002 


1.134 


0.1 32 


A ft ft 

1 3.2 


SMALL E. CITIES, MO 


1 1260 


ft ft 

02 


0.649 


ft "7"7 

0.778 


ft A ft ft 

0.129 


a n n 

1 9.9 


P\ A | 1 AO TV 

DALLAS, TX 


r\ r\ r\ r\ r\ 

00900 


A A 

1 1 


0.863 


ft ft ft ft 

0.989 


ft A OC 

0.126 


-1 A C 

14. D 


rvrklTAM T\/ 

DENTON, TX 


ft ft ft ft ft 

00900 


A ft 

12 


0.863 


ft ftOft 

0.989 


ft A ft ft 

0.126 


14. D 


POUGHKPSIE/N.NYC SUBURBS 


00803 


03 


1.104 


1.184 


0.080 


7.3 


D.C. + MDA/A SUBURBS 


00580 


01 


1.374 


1.452 


0.078 


5.7 


KHUUE loLAlNU 


AA07A 

00870 


r\A 
01 


A ft CO 

1 .058 


A A ft A 

1 .1o4 


U.U/D 


7 1 


oUU 1 H + E. SHORE MD 


ftftftr\r\ 
00690 


03 


a ft a r 

1 .01 5 


A ftQE 

1 .085 


n r\~7r\ 
U.U /L) 


b.y 


MADTUCDM Kit — \ n / II — n CM — \/ 

NORTHERN NEW JERSEY 


ftftOCft 

00860 


ft A 

01 


A ft ft ft 

1 .339 


A A no 

1 .408 


o.uby 


0.2 


Kill a aai cri 

MIAMI, FL 


ft ft c. C\(~\ 

00590 


ft A 

04 


1 .071 


1 .1 39 


U.Ubo 




VERMONT 


AAlfl ft 

00780 


f ft 

50 


0.965 


A ft ft ft 

1 .032 


0.067 


b.9 


SOUTHERN MAINE 


ft A ft ft ft 

21200 


03 


1 .080 


1 .143 


ft ft ft ft 

0.063 


5.8 


PHILLY/PITT MED SCHS/HOSPS 


00865 


i 


0.985 


1.043 


ft ft c o 

0.058 


5.9 


DELAWARE 


00570 


01 


1.051 


1 .106 


0.055 


5.2 


OM T/""\\ A #K 1 /I K IT » IPTniAl \/A 

SM. TOWN/IND JSTRIAL VA 


A ft A ft ft 

10490 


ftft 

03 


ft "7 ft ft 

0.763 


ft. O A 1 

0.81 7 


r\ nest 
U.U54 




SANTA BARBARA, CA 


02050 


16 


1.247 


1.298 


0.051 


4.1 


Largest Decreases 














RENO, ET AL (CITIES), NV 


01290 


02 


1.427 


1.271 


-0.156 


A A ft 

-1 1 .0 


PHOENIX, AZ 


01030 


01 


1.134 


0.981 


-0.153 


Aft r~ 

-13.5 


OAKLAND-BERKELEY, CA 


00542 


07 


1.580 


1.441 


-0.139 


-8.8 


SANTA CLARA, CA 


00542 


09 


1.719 


1.593 


-0.126 


~7 ft 

-7.3 


NEW ORLEANS, LA 


00528 


01 


1 .01 1 


0.891 


-0.120 


4 -i n 
-1 1 .9 


DETROIT, Ml 


00710 


01 


1.053 


0.938 


-0.115 


-10.9 


UTAH 


00910 


09 


0.917 


0.804 


-0.113 


-12.3 


SPOKANE+RICHLND(CITIES) 


00932 


03 


0.970 


0.863 


-0.107 


-11.0 


ALASKA 


01020 


01 


1.304 


1.197 


-0.107 


-8.2 


E.CEN+NE WA (EXCL SPOKANE) 


00932 


04 


0.932 


0.830 


-0.102 


-10.9 


ABILENE, TX 


00900 


29 


0.849 


0.755 


-0.094 


-11.1 


TUCSON (CITY), AZ 


01030 


02 


1.084 


0.992 


-0.092 


-8.5 


SAN ANGELO, TX 


00900 


30 


0.855 


0.771 


-0.084 


-9.8 


OKLAHOMA 


01370 


00 


0.799 


0.718 


-0.081 


-10.1 


SOUTHEAST IL 


00621 


13 


0.751 


0.681 


-0.070 


-9.3 


QUINCY, IL 


00621 


07 


0.751 


0.693 


-0.058 


-7.7 



II-3-11 



gpci\rentrep\TAB3-3.XLS 



2 



TABLE 3-3 

LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE RENTAL INDEX AS A RESULT OF USING 
A BLEND OF 1991, 1992, AND 1993 FAIR MARKET RENTS (FMRs) (a) 









1987 


, 91,'92, , 93 




Percent 


Locality Name 


Carrier 


Locality 


FMR 


FMR 


Change 


Change 


TEMPLE, TX 


00900 


06 


0.773 


0.718 


-0.055 


-7.1 


COLORADO 


00550 


01 


0.981 


0.927 


-0.054 


-5.5 


N. K.C. (CLAY/PLATTE), MO 


00740 


02 


0.885 


0.834 


-0.051 


-5.7 


K.C. (JACKSON COUNTY), MO 


00740 


03 


0.885 


0.834 


-0.051 


-5.7 


SUBURBAN KANSAS CITY, KS 


00740 


04 


0.885 


0.834 


-0.051 


-5.7 


KANSAS CITY, KS 


00740 


05 


0.885 


0.834 


-0.051 


-5.7 


SOUTHEAST (OHIO VALLEY) OH 


16360 


15 


0.769 


0.718 


-0.051 


-6.6 



(a) More than a 5 percentage point gain or loss. 



II-3-12 



gpci\rentrep\TAB3-3.XLS 



TABLE 3-4 



LOCALITIES EXPERIENCING LARGE CHANGES IN OFFICE RENTAL INDEX AS A RESULT OF USING 
COUNTY-SPECIFIC RENTS, 1993 1 



1993 









1993 


County-Specific 




Percent 


Localitv Name 


Carrier 


Localitv 


FMR 2 


FMR 2 


Chanqe 


Chanqe 


Largest Increases 














MANHATTAN, NY 


00803 


01 


1.628 


2.102 


0.475 


29.2 % 


SUBURBAN CHICAGO, IL 


00621 


15 


1.243 


1.337 


0.094 


7.6 


QUEENS, NY 


14330 


14 


1.628 


1.705 


0.077 


4.7 


MILWAUKEE SUBURBS, Wl (SE) 


00951 


46 


0.897 


0.952 


0.055 


6.2 


EASTERN CONN. 


10230 


04 


1.156 


1.207 


0.051 


4.4 


Largest Decreases 














NYC SUBURBS/LONG I., NY 


00803 


02 


1.615 


1.476 


-0.138 


-8.6 


PHILLY/PITT MED SHCS/HOSPS 


00865 


01 


1.050 


0.965 


-0.085 


-8.1 



1 More than a 5 percentage point gain or loss. 
2 NYC FMR replaced by Bergen-Passaic FMR. 



II-3-13 



gpci\pracexp\rent\TAB3-4.XLS 



comprised of the other boroughs of New York City, Long Island, and the northern New York 
City suburbs loses 9 percent. 

3.3.4 Two Bedroom Versus Four Bedroom FMRs 

Table 3-5 shows localities with the largest differences between GPCI rental indices 
computed from the 2 and 4 bedroom FY 94 FMRs (published by HUD in the October 1, 1993 
Federal Register) . The differences are non-trivial, but not large. The two indices differ by 
less than 10 percent in all areas. Since the rental index comprises about 10 percent of the 
overall Geographic Adjustment Factor (GAF), the effect of this choice on the GAF is less than 
one percent everywhere. There is no readily discernible pattern to the differences. 



gpci\ piacexp\ rent\chap>3 



II-3-14 



1 



TABLE 3-5 



LOCALITIES WITH THE LARGEST DIFFERENCES BETWEEN OFFICE RENTAL INDICES BASED ON 
TWO- AND FOUR-BEDROOM FAIR MARKET RENTS (FMRs) 









1994 


1994 












2 Bedroom 


4 Bedroom 




Percent 


Locality Name 


Carrier 


Locality 


FMR 


FMR 


Difference 


Difference 


Largest Increases 














MANHATTAN, NY 


00803 


01 


1.842 


1.722 


0.120 


7.0 % 


QUEENS, NY 


14330 


04 


1.494 


1.396 


0.098 


7.0 


SAN DIEGO/IMPERIAL, CA 


02050 


28 


1.253 


1.172 


0.081 


6.9 


SUBURBAN CHICAGO, IL 


00621 


15 


1.291 


1.212 


0.078 


6.5 


VIRGIN ISLANDS 


00973 


50 


1.096 


1.023 


0.073 


7.1 


CHICAGO, IL 


00621 


16 


1.172 


1.100 


0.071 


6.5 


BAKERSFIELD, CA 


00542 


14 


1.077 


1.007 


0.070 


6.9 


DENTON, TX 


00900 


12 


1.015 


0.946 


0.068 


7.2 


ALASKA 


01020 


01 


1.221 


1.153 


0.068 


5.9 


DALLAS, TX 


00900 


11 


0.999 


0.931 


0.067 


7.2 


ROCHESTER/SURR. CNTYS, NY 


00801 


02 


0.993 


0.927 


0.066 


7.1 


FRESNO/MADERA, CA 


00542 


11 


0.985 


0.920 


0.065 


7.0 


ROCKFORD, IL 


00621 


02 


1.022 


0.957 


0.065 


6.8 


URBAN MASS 


00700 


01 


1.333 


1.269 


0.065 


5.1 


VICTORIA, TX 


00900 


32 


1.033 


0.969 


0.065 


6.7 


MILWAUKEE SUBURBS (SE), Wl 


00951 


46 


0.992 


0.927 


0.065 


7.0 


ODESSA, TX 


00900 


13 


0.958 


0.896 


0.063 


7.0 


MIDLAND, TX 


00900 


23 


0.958 


0.896 


0.063 


7.0 


DETROIT, Ml 


00710 


01 


1.003 


0.941 


0.062 


6.6 


AUSTIN, TX 


00900 


31 


0.9?8 


0.876 


0.062 


7.1 


MILWAUKEE, Wl 


00951 


04 


0.913 


0.851 


0.061 


7.2 


PEORIA, IL 


00621 


05 


0.940 


0.879 


0.060 


6.9 


YUMA, AZ 


01030 


08 


0.922 


0.863 


0.059 


6.8 


SAN FRANCISCO, CA 


00542 


05 


1.760 


1.704 


0.057 


3.3 


MASS SUBURBS/RURAL CITIES 


00700 


02 


1.209 


1.152 


0.057 


4.9 


N. CENTRAL CITIES, NY 


00801 


03 


0.925 


0.868 


0.057 


6.6 


BATON ROUGE, LA 


00528 


03 


0.878 


0.822 


0.056 


6.8 


SAN MATEO, CA 


00542 


06 


1.742 


1.686 


0.056 


3.3 


METROPOLITAN, IN 


00630 


01 


0.850 


0.794 


0.056 


7.1 


TUSCON, AZ 


01030 


02 


0.854 


0.798 


0.056 


7.0 


MERCED/SURR. CNTYS, CA 


00542 


10 


1.003 


0.948 


0.055 


5.8 


LAFAYETTE, LA 


00528 


06 


0.836 


0.782 


0.054 


6.9 


DES MOINES (POLK/WARREN), IA 


00640 


05 


0.887 


0.833 


0.054 


6.5 


SAN ANTONIO, TX 


00900 


07 


0.824 


0.770 


0.054 


7.1 


LONGVIEW, TX 


00900 


17 


0.836 


0.783 


0.054 


6.9 


CORPUS CHRISTI, TX 


00900 


24 


0.870 


0.815 


0.054 


6.7 


ROCK ISLAND, IL 


00621 


04 


0.837 


0.784 


0.053 


6.7 


SOUTHERN MAINE 


21200 


03 


1.122 


1.070 


0.053 


4.9 



II-3-15 



gpci\pracexp\rent\TAB3-5.XLS 



2 



TABLE 3-5 

LOCALITIES WITH THE LARGEST DIFFERENCES BETWEEN OFFICE RENTAL INDICES BASED ON 
TWO- AND FOUR-BEDROOM FAIR MARKET RENTS (FMRs) 



1994 1994 
2 Bedroom 4 Bedroom Percent 



Locality Name 


Carrier 


Locality 


run 

FMR 


FMR 


Difference 


Differen< 


VVAOU, I A 


00900 


22 


a ~~7 O A 

0.784 


A 7rtO 

0.733 


A A C A 

0.052 


/.o 


ni iccai r\/c[ idd mtvc mv 
DUrrALU/oUKK. UN I Yo, NY 


00801 


01 


A Q07 

0.83/ 


a 7oc 

0.786 


A A C H 

0.051 


D.O 


MININbSOT A (TRAVbLbRS) 


00720 


Art 

00 


A AO A 

0.933 


a o o o 

0.883 


A AC A 

0.050 


b. / 


LAKbDO IX 


00900 


33 


0.775 


0.726 


A ACA 

0.050 


o.y 


LARGEST DECREASES 














Uu +MD/VA SUBURBS 


00580 


01 


H C" A A 

1 .529 


•1 C7C 

1 .675 


A A A C 

-0.146 


-0. / 


a a i/i a Mr\/Drni/i r— \/ /-\ a 

OAKLAND/BERKLEY, CA 


00542 


07 


1 .420 


-4 r r r 

1 .555 


A HOC 

-0.135 


Q "7 

-0. / 


LOS ANGELES (1ST OF 8) 


02050 


18 


1 .506 


1 .618 


A A A A 

-0.1 12 


-o.y 


LOS ANGELES (2ND OF 8) 


02050 


19 


1 .506 


1 .618 


A A A A 

-0.1 12 


-o.y 


LOS ANGELES (3RD OF 8) 


02050 


20 


1 .506 


1.618 


A H A A 

-0.1 12 


-o.y 


LOS ANGELES (4TH OF 8) 


02050 


21 


1 .506 


1 .618 


-0.1 12 


-6.9 


LOS ANGELES (57 H OF 8) 


02050 


22 


1 .506 


1 .618 


-0.1 12 


-6.9 


I /*"\ o am r"~ i r - /ATt i /"s r - o \ 

LOS ANGELES (6TH OF 8) 


02050 


23 


1 .506 


1 .618 


A A A A 

-0.1 1 2 


-b.y 


LOS ANGELES (7TH OF 8) 


02050 


24 


1 .506 


1 .618 


A H A A 

-0.1 12 


-6.y 


LOS ANGELES (8TH OF 8) 


02050 


25 


1.506 


1 .618 


-0.1 12 


-6.9 


SEATTLE (KING CNTY), WA 


01390 


02 


1.157 


1.268 


-0.1 11 


O "7 

-8.7 


LAS VEGAS, ET AL. (CITIES), NV 


01290 


01 


1.077 


1 . 1 79 


-0.102 


-8.7 


SACRAMENTO/SURR. CNTYS, CA 


00542 


04 


1.087 


1.188 


-0.101 


-8.5 


MADISON (DANE COUNTY), Wl 


00951 


15 


1.032 


1.129 


-0.098 


-8.6 


TIDEWATFR & N VA CNTYS 


10490 


02 


1.060 


1.157 


-0.097 


-8.4 


HOUSTON, TX 


00900 


18 


0.970 


1.062 


-0.092 


-8.6 


BRAZORIA, TX 


00900 


09 


0.964 


1 .055 


A A A 

-0.0»1 


-O.D 


FORT WORTH, TX 


00900 


28 


0.930 


1 .019 


-0.089 


-8. / 


GALVESTON, TX 


00900 


15 


0.903 


0.987 


-0.085 


-8.6 


N. COASTAL CNTYS, CA 


00542 


01 


1.220 


1.304 


-0.084 


-6.5 


PHOENIX, AZ 


01030 


01 


0.892 


0.976 


-0.084 


-8.6 


MIAMI, FL 


00590 


04 


1.286 


1.368 


-0.082 


-6.0 


NEW ORLEANS, LA 


00528 


01 


0.802 


0.879 


-0.078 


-8.8 


EL PASO, TX 


00900 


14 


0.810 


0.887 


-0.077 


-8.7 


ATLANTA, GA 


01040 


01 


1.021 


1.098 


-0.076 


-7.0 


RICHMOND & CHARLOTTESVILLE, VA 


10490 


01 


0.932 


1.007 


-0.075 


-7.5 


FORT LAUDERDALE, FL 


00590 


03 


1.167 


1.241 


-0.074 


-5.9 


MIDDLE. NJ 


00860 


02 


1.442 


1.515 


-0.074 


-4.9 


FLAGSTAFF, AZ 


01030 


05 


0.972 


1.046 


-0.073 


-7.0 


REST OF NEVADA 


01290 


99 


1.046 


1.116 


-0.070 


-6.2 


SHREVEPORT, LA 


00528 


02 


0.730 


0.799 


-0.069 


-8.6 


STOCKTON/SURR. CNTYS, CA 


00542 


08 


1.005 


1.074 


-0.069 


-6.4 


LAKE CHARLES, LA 


00528 


04 


0.732 


0.800 


-0.068 


-8.5 



II-3-16 



gpci\pracexp\rent\TAB3-5.XLS 



3 



TABLE 3-5 



LOCALITIES WITH THE LARGEST DIFFERENCES BETWEEN OFFICE RENTAL INDICES BASED ON 
TWO- AND FOUR-BEDROOM FAIR MARKET RENTS (FMRs) 









1994 


1994 












2 Bedroom 


4 Bedroom 




Percent 


Locality Name 


Carrier 


Locality 


FMR 


FMR 


Difference 


Difference 


AMARILLO, TX 


00900 


26 


0.702 


0.770 


-0.068 


-8.8 


COLORADO 


00550 


01 


0.909 


0.975 


-0.066 


-6.7 


WICHITA FALLS, TX 


00900 


34 


0.690 


0.756 


-0.066 


-8.7 


ABILENE, TX 


00900 


29 


0.674 


0.737 


-0.063 


-8.5 


EUGENE, ETAL. (CITIES), OR 


01380 


02 


0.897 


0.957 


-0.060 


-6.2 


ELKO & ELY (CITIES), NV 


01290 


03 


0.989 


1.049 


-0.059 


-5.7 


MARIN/NAPA/SOLANO, CA 


00542 


03 


1.399 


1.457 


-0.058 


-4.0 


KINGS/TULARE, CA 


00542 


13 


0.834 


0.892 


-0.058 


-6.5 


VENTURA, CA 


02050 


17 


1.584 


1.639 


-0.055 


-3.3 


ANAHEIM/SANTA ANA, CA 


02050 


26 


1.537 


1.590 


-0.053 


-3.3 



II-3-17 



gpci\pracexp\rent\TAB3-5.XLS 



4.0 



NEW YORK CITY AREA OFFICE RENTAL INDICES 



The office rental index for New York City is problematic. New York City rent control 
ordinances may lower residential apartment rents below free market levels (see Chapter 2), 
but rent control does not apply to commercial office space. Thus, the New York City FMR 
may not be a good indication of relative physician office rents in New York City. Because of 
this problem, the New York City FMR was replaced by the FMR of Bergen-Passaic, New 
Jersey in computation of the current Medicare Fee Schedule office rental index. The Bergen- 
Passaic FMR was chosen because it was the highest in the New York City area CMSA, and 
spatial economic theory suggests that rents in the urban core will be at least as high as 
elsewhere in the metropolitan area. 

An additional limitation of the current New York City office rental index is that it is 
the same for all counties in the New York City PMSA. In reality, rents may vary 
significantly within New York City. Manhattan physicians have claimed that their office 
rents are substantially higher than in other boroughs. If New York were a single Medicare 
payment locality, an average city-wide rent would suffice to establish an accurate office 
rental index. But because Manhattan and Queens are distinct Medicare payment localities, 
measuring rental variation by borough is necessary to create appropriate rental indices for 
the New York City localities. 

Thus, two significant issues surrounding the New York City office rental indices must 
be resolved: 

(1) What is the appropriate office rental index for New York City, on average? 

(2) Do rents vary by borough within Ne-*' York City? If so, how can th be 
reflected in the office rental index? 

This chapter considers both of these issues. First, in Section 4.1, we evaluate evidence from 
surveys of physician practices on office expense per square foot in the New York City area. 
Then, in Section 4.2, we discuss evidence on apartment rents in the New York City PMSA. 
Section 4.3 shows the impact of incorporating county-specific rents and alternative average 
New York Cily rental values on rental indices for New York City area Medicare payment 
localities. 



gpci\ pracexp\ rent\ chap4 



II-4-1 



4.1 Evidence From Physician Surveys 

Two surveys of physician practices collect information on physician office expenses 
per square foot: the Health Care Financing Administration's Physician Practice Costs and 
Income Survey (PPCIS) and the American Medical Association's (AMA) Socioeconomic 
Monitoring Survey (SMS). PPCIS rents include utilities and telephone in addition to annual 
tax deductible rental, lease, or depreciation/ interest costs for space. (The PPCIS rental data 
are analyzed in greater detail in Chapter 5.) The SMS requests "office expenses," including 
annual rent, lease, or mortgage expenses. We we f e able to identify PPCIS physicians by 
county, but SMS physicians only by Medicare locality. We put office expenses from each 
survey on a square foot basis, then calculated median expense per square foot nationally and 
for New York City area counties or payment localities. The New York medians were 
standardized by the national median cost per square foot to create an index analogous to the 
GPCI office rental index. PPCIS rents are for 1988, and SMS rents are for 1991 or 1990. 

Table 4-1 shows PPCIS rental indexes for the New York City area, together with the 
number of physician practices on which they are based (N). Approximate 90 percent 
confidence intervals for the indexes were calculated using a nonparametric statistical 
procedure (DeGroot, 1975, p. 472). The PPCIS office expense index for the New York PMSA 
is 1.38, which is less than its value of 1.578 in the current (FY 1987 FMR) GPCI office rental 
index or its value of 1.628 if Bergen-Passaic's fiscal year 1993 FMR is substituted for the New 
York rMR. However, the confidence interval for the index value is rather large because it is 
based on only 68 practices. There is a 90 percent chance that New York's "true" index value 
is between 1.22 and 1.63. This range includes the office rental indices implied by the Bergen- 
Passaic substitution, so they cannot be statistically rejected. However, the Bergen-Passaic 
substitution does put the New York index at the upper limit of the 90 percent confidence 
interval. 

We also computed index values for Manhattan and for the other four New York City 
boroughs combined (Queens, Bronx, Brooklyn, and Staten Island). Both these indexes are 
estimated quite imprecisely from a small number of practices, as is shown by their enormous 
confidence intervals. Nevertheless, by a formal statistical test, Manhattan's office rent is 
statistically greater than that of the other four boroughs at the 90 percent level of confidence. 
The estimated index for Manhattan is 2.23 compared to 1.04 for the other boroughs. Not 
much weight should be put on these particular index values because of the considerable 
imprecision with which they are estimated. 



gpd\ pracexp\ rent\ chap4 



II-4-2 



TABLE 4-1 



NEW YORK CITY AREA MEDIAN PHYSICIAN OFFICE COST PER SQUARE FOOT: 
PHYSICIAN PRACTICE COST AND INCOME SURVEY, 1988 
(indexed to national median cost per square foot = 1 .0) 



Confidence Significance 
Area Index Interval Level 3 N 



NYCPMSA 1.38 1.22, 1.63 89 % 68 

NYC 1.40 1.13, 1.63 89 57 

Manhattan 2.23 1.40, 2.68 91 22 

Other Boroughs 1.04 0.88, 1.59 90 35 



a Of confidence interval. 

NOTE: Includes utilities and telephone in addition to rent. 
SOURCE: 1988 Physician Practice Cost and Income Survey. 



II-4-3 



gpci\pracexp\rent\TAB4-1 .XLS 



The PPCIS data thus indicate that: (1) the Bergen-Passaic FMR provides an office 
rental index for New York City that is in the acceptable range, but is at the upper limit of 
that range; (2) Manhattan rents are greater than those in the other four boroughs. Table 4-2 
shows office expense indexes from the 1991 and 1990 AMA surveys by Medicare payment 
locality. (We did not calculate confidence intervals for the AMA indexes, but they are 
probably similar to the PPCIS intervals for a similar number of observations.) The AMA 
indexes are consistent with the PPCIS indexes. They indicate that a reasonable office rental 
index for the New York City area is 1.5 to 1.6, and that Manhattan rents may be higher than 
those in other New York City area localities. 

4.2 Apartment Rents by County in the New York Metropolitan Area 

Apartment rents alone may not be able to establish appropriate office rental indices 
for the New York City area because rent control may cap them below market levels. 
However, relative apartment rents may be useful in determining relative office rental indices 
for counties within the New York City PSMA. HUD does not produce FMRs for individual 
metropolitan counties. To measure county-specific rents, we obtained a HUD special 
tabulation of 1990 Census data. We utilize the 45th percentile of the rental distribution of 
two-bedroom apartments rented by recent movers to be consistent with the FMRs. 

These rents are shown for all counties in the New York City PMSA in Table 4-3. 
Westchester county, New York is excluded because HUD produces a distinct FMR for 
Westchester county. The Census rents show a large variation among counties. Manhattan is 
the highest, at $931 per month, followed by Putnam and Rockland counties. Rents in the 
Bronx and Brooklyn, at $537 and $635, respectively, are much lower. These differences are 
based on sample sizes of thousands of apartments and are highly statistically significant. The 
Census apartment rents, then, provide strong evidence of rental variation among New York 
City counties. 

4.3 Office Rental Indices for New York City Area Localities 

The evidence from surveys of physician practice costs and from the Census apartment 
rents indicates significant variation in rents among counties in the New York City PMSA. 
This variation in physician costs should be reflected in the GPCI office rental index. Because 
the Census apartment rents are consistent with the HUD FMRs, which are the basis of the 

gpti\pracexp\rent\chap>4 1 1 -4-4 



TABLE 4-2 



NEW YORK CITY AREA MEDIAN PHYSICIAN OFFICE EXPENSES PER SQUARE FOOT: 
AMA DATA, 1991, 1990 

(indexed to national median cost per square foot = 1 .0) 



1991 



1990 



Locality 

All "NYC" localities 8 
Manhattan 

Queens, other boroughs.suburbs, 
Long Island 



Index 



N 



1.51 76 
1.61 21 
1.50 55 



Index 

1.60 
2.93 
1.52 



N 

39 
14 
25 



'Manhattan, Queens, other boroughs, suburbs, Long Island. 

Source: American Medical Association Socioeconomic Monitoring Survey. 



II-4-5 



gpci\pracexp\rent\TAB4-2.XLS 



TABLE 4-3 



COUNTY-SPECIFIC APARTMENT RENTS FOR THE NEW YORK CITY PMSA 1 



45th 
Percentile 



Countv 


Rent 2 


Manhattan 


$931 


Queens 


$755 


Staten Island 


$715 


Brooklyn 


$635 


Bronx 


$537 


Putnam 


$893 


Rockland 


$837 



'Excludes Westchester county, for which HUD calculates a separate fair market rent (FMR). 

'Gross monthly rent, 2-bedroom apartments, recent movers, from HUD special tabulation of the 1990 Census. 

SOURCE: HUD Special Tabulation of 1990 Census data. 



II-4-6 



gpcrpracexp\rent\TAB4-3.XLS 



office rental index, they can be used to define relative county -specific rents for the office 
rental index. 

In Table 4-4, we present options for updated office rental indices for New York City 
area Medicare payment localities. 1 All the options reflect the adjustment of HUD's FMR by 
county-specific rents for each New York City borough (and for all other counties in the New 
York CMSA) from a special tabulation of 1990 Census rental data. The six options for 
updating are: 

1. Use the actual FY 94 FMR for New York City. 

2. Increase New York City's FMR by 10 peicent to offset the effects of residential 
rent control in New York City. The 10 percent adjustment is derived from a 
study of the amount by which rent control lowers New York City rents cited 
in HUD's publication Report to Congress on Rent Control, September 1991. 

3. Replace New York City's FMR with the FMR of Bergen-Passaic, New Jersey. 
This is an update of what is currently being done. 

4. Replace New York City's FMR with the FMR of Westchester county, New 
York. Even though Westchester county is in the same primary MSA as New 
York City, HUD is required by Congress to establish a separate FMR for 
Westchester county. 

5. Replace New York City's FMR with the average of the New York City area 
CMSA, excluding New York City itself. 

6. Replace New York City's FMR with the average of the New York City area 
CMSA, including New York City. 

Table 4-4 shows the current rental index and the six updating options for New York 
City area Medicare payment localities. The locality labelled "NYC suburbs/ Long I., NY" 
includes the New York City boroughs of Brooklyn, the Bronx, and Staten Island, in addition 
to Nassau and Suffolk counties on Long Island, and Westchester and Rockland counties 
north of New York City. The rental indices of nearby localities and of selected other high- 
rent localities across the country are also shown for comparison. The comparison values are 
not repeated for each New York City option, since they change only slightly. 

Compared to its current (1992) rental index, Manhattan's rental index rises under the 
first updating option, using the actual New York City FMR. Conversely, the rental indices of 
Queens and "NYC Suburbs" fall. This is due to the adjustment of HUD's New York City 
FMR by county-specihc rents within the New York City metropolitan area. Manhattan rents 
are much higher than rents in the other New York City boroughs. Under the actual FMR 



The indices shown in Table 4-4 are based on the preliminary version of the final FY 1994 FMRs published in the October 
1, 1993 Federal Registe r. The final office rental index used for payment is based on the final FY 1994 FMRs published by HUD 
in the April 6, 1994 Federal Register . 



gpci\ pracexpX rent\ chap4 



II-4-7 



CO 
UJ 

o 

o 

z 

-I 
< 

Z 
HI 

Q£ 

Q 
LU 
I- 
< 
Q 

a. 

3 



O a) 

£ £ o 

z < >■ 

a> < Z 

£ W - 

*^ 

a> < c 

V V"* o 

Q£ > X 
5 uj 



-I 

£z J 



Is 

p 
2 5 





CO 


to 




CM 


to 




•05 




lO 



















CD 




co 


o 




& 




CD 




CM 













ID 






h« 


CD 


: 


a 


-= 


(-- 


to 




CM 




T — 


™ 



O 


CO 


OJ 


c , 


'■! 


CO 


CM 




O 


CD 


lO 


CM 


CM 









I 

LU 

I 

CD 

f 



CO 
LU 

g 

a 
z 

_j 

f 

z 

Hi 
UJ 

o 

LL 
LL 
O 

2 

< 



O 

a: 
o 
>- 



LU 
Z 

Q 
LU 
I- 
< 
Q 
0. 
3 

£ 
O 

LL 
CO 



£L 
O 



cu o 

(A 

z 



S «J x 

3 5 C 



o 



m 



CO 
LU 



< 

o 
o 



>- 
z 







>- 



2£ 



O 
>- 



>- 

z 

co" 

CD 
0. 
3 
00 
3 
00 

O 
rao UJ 

* co 

00 X 
3 X 
LU W O 
■ O =) 
>- O 

z a. 



LU 
3 

o 



co 


o 


OO 


CO 


























CO 


CM 


o 
























o 


CO 


m 


CM 
























CM 






T— 
























CM 




CO 


oo 


oo 


CM 


o 


co 


o 


CM 


o> 


CD 


CO 




CM 




Oi 


CO 


00 


CO 




CO 








CM 


o 


CO 


a> 




00 






CM 


■<fr 




CD 








m 


in 


CO 


CM 














T — 














T~" 


T — 




00 


00 


m 




O) 


00 




oo 


o 


o 




CD 






CO 




h- 


CO 


o 


CO 


■<* 




CD 


CO 


CO 










OJ 


iq 


in 


in 




co 


CM 


CO 


CM 


h-. 




CO 






CM 


























•< — 


■"! — 










CM 


CO 


T — 


CM 


CM 




«n 


CD 






T — 


m 


CD 


o 


o 


o 


o 


o 


o 


o 


o 


o 


o 


o 


CM 


o 




T — 
























i 

OO 






































00 


o 


CO 


CO 


o 


o 


o 


o 


CM 


CM 


o 


o 


o 




-= 


o 


CO 


o 


o 


co 


co 


CO 


CM 






00 


m 


o 


CM 


CM 


CO 


co 


co 


oo 


oo 


00 


CM 




m 


m 


m 


o 


1^- 


CD 


©■ 


o 




o 


o 


o 


o 


o 




o 


o 


o 


CM 


o 


o 


O 


o 


T — 


o 


o 


o 


o 




o 


o 


o 


o 


o 


o 


o 


o 



co 

LU 



< 

o 
o 

-1 
>- 

m 



s 
Z 



DC 
LU 
X 
H 

a: 
o 



z 

UJ 

_J 
Q 
Q 



(0 
LU 



-1 
< 
O 
C 
-I 

I- 

z 

LU 
Of 

X 

o 



< 
o 

6 

CO 

o 
o 



o 



CO 



LU < LL 

X ^ Z 

t < < 

O X CO 



CO 
CD 
01 

CD 

3 CO 
CO LU 



LU 
CD 

z 
< 

o 



o 
o 
< 
o 

X 

CO O 

w z 
< 

QJ 



< 

CD CD — 



CO O 



II-4-R 



option, Manhattan's rental index is the highest in the country, slightly exceeding the rental 
indexes of Hawaii and San Francisco. The rental indices of Queens and "NYC Suburbs" are 
about equal to those of Northern and Middle New Jersey, but less than Southwest 
Connecticut. 

The various options that adjust or replace the actual FMR raise the rental indices of 
Manhattan and Queens by 4 percent (replace NYC with CMSA average including NYC) to 18 
percent (replace NYC with Westchester). The "NYC Suburbs" rental index is increased by a 
smaller percentage in these options. Three of the options -- increase NYC by 10 percent, 
replace NYC with Bergen-Passaic, and replace NYC with CMSA average excluding NYC — 
result in very similar index values. Each of these options would raise the overall Geographic 
Adjustment Factor (GAF) of Manhattan and Queens by about 1 percent compared to use of 
the actual FMR, because the rental index accounts for about 10 percent of the overall GAF. 



gpci\ pracexp\ rent\ chap4 



II-4-3 



5.0 VALIDATION OF OFFICE RENTAL INDEX USING PHYSICIAN SURVEY DATA 



The office rental index cannot currently be based upon actual physician expenditures 
for office space, since no survey exists providing data on this expense for a sufficiently large 
sample. However, survey data on physician expenditures can be used to examine the extent 
to which the index reflects variations in actual cos Is for office space incurred by physicians. 
Two previous studies comparing the GPCI index with input price data from the American 
Medical Association are discussed in section 5.1. Section 5.2 presents our results comparing 
the GPCI index with physician expenditure data ; iom another source—HCFA's most recent 
Physicians' Practice Costs and Income Survey 

5.1 Previous Validations 

Studies by Gillis et al. (1992) and Zuckerman et al. (1992) have utilized data from the 
American Medical Association's Socioeconomic Monitoring System (SMS) survey to compare 
variation in office costs reflected by the GPCI with those reported by physicians. The SMS is 
a nationally representative survey of non-federal patient care physicians (excluding residents) 
stratified by specialty and census region. 1 Self-employed physicians are asked to report their 
share of practice expenses for six expense categories: professional liability premiums, 
nonphysician employee payroll, office expenses, medical equipment, medical supplies, and 
miscellaneous other expenses. Assuming that resources were divided equally among all 
owners of the practice, cost per square foot of office space can be calculated by dividing the 
reported cost for office expenses by the number of square feet of office space per physician 
owner in the practice. Using geographic identifiers provided by the AMA, the office rental 
index can be merged onto this file for each observation. 

Both studies use data on office costs reported on the SMS for both 1990 and 1991. 
Gillis et al. rely solely on the GPCI values reported in the June 5, 1991 Federal Register which 
were constructed using HUD Fair Market Rent data for 1987. Thus, some discrepancies 
between reported physician prices and GPCI values may reflect changes in the relative costs 
of office space occurring during the late 1980's. Zuckerman et al. supplement analysis of the 



'Gonzales (1992) provides a detailed discussion of the survey methodology. 



gpci\pracexp\rent\chap5 



II-5-1 



1987 FMRs with more recent data from 1990 to determine whether this difference in survey 
year affects the results. 

Gillis et al. and Zuckerman et al. both begin by testing whether there were significant 
geographic variations in the cost per square foot of office space reported by physicians. A 
finding of no geographic variations in cost would imply that the office rental component of 
physician expenses could be treated as a national market, and no GPCI adjustment would be 
necessary. 

Specifically, they estimate the regression: 

In (COSTy) = lna + P'X ij + C j D j + E Sj (1) 

where COST^ is the cost per square foot of office space reported by physician i practicing in 
locality j; Xjj is a vector of physician characteristics; and Dj is a dummy variable defined to be 
one if the physician practices in locality j and zero otherwise. 

The joint F-test on the set of locality dummy coefficients provides a test of the null 
hypothesis that the input prices are identical across all localities. In both studies, this 
hypothesis is rejected at the 1 percent significance level, implying that input prices do vary 
across localities. 

Second, each study tests whether the input price (cost per square foot) and the GPCI 
were correlated across areas by estimating the model: 

In (COST^) = lna + p' Xq + yln GPCI ( + E , (2) 

where COST^ and X tj are defined as above, and In GPCIj is the natural log of the GPCI for 
locality j. In this specification the coefficient measures the relationship between reported 
expenditures per square foot and the GPCI. A value of zero would indicate no relationship 
between the GPCI and the input price; a value of one would indicate the GPCIs and the 
input prices moved proportionately. In economic terminology, y represents the elasticity of 
the SMS cost data with respect to the GPCI. An elasticity of one would imply, for example, 
that a 10% increase in input prices corresponded to a 10% increase in the GPCI. Gillis et al. 
estimate this regression using three specifications for the vector of physician characteristics: 
year dummies only; office type and year dummies; and specialty, office type and year 
dummies. They also estimate the model for a subsample consisting of only solo practitioners. 
The estimated coefficients on ln(GPCI) are relatively insensitive to model specification, 



gpci\ pracexp\ rent\chap5 



II-5-2 



ranging from .70 to .77. These estimates were all significantly greater than zero, however. 
All were also significantly less than one. Zuckerman et at estimate the regression using 
GPCI data calculated using Fair Market Rent data from three time periods and find 
coefficients ranging from .56 to .62. These results imply that while the GPCIs and SMS data 
are positively correlated, the SMS has significantly less variation across observations than 
does the GPCI. 

Gillis et al. undertake two other tests for systematic variation in the GPCIs. First, they 
deflate the natural log of cost per square foot by natural log of the locality GPCI, and regress 
this ratio on a set of locality dummy variables. If the GPCI perfectly reflected variation in 
the input price the ratio between the two would equal one. If there were no systematic 
variation in the relationship between the GPCI and the price estimates, the coefficients on the 
locality dummy variables would not be significantly different from zero. The joint test that 
the GPCI-adjusted prices were the same across all localities was rejected. 

To characterize areas that appear to be over- or under- compensated by the GPCI, the 
analysis of GPCI-adjusted input prices was repeated with the locality dummy variables 
replaced by a set of location dummy variables. These specify the county where the 
physician's practice was located as urban or rural, with separate classes for population 
categories. In this specification, the joint test for significance of the location dummies implies 
there is no systematic variation using this stratification. 

5.2 Validation using the PPCLS 
5.2.1 Data Description 

The data source for this analysis is the 1988 Physicians' Practice Costs and Income 
Survey (PPCIS), sponsored by the Health Care Financing Administration. To be eligible for 
the survey, a physician was required to provide patient care services (excluding residents) 
and not be employed by a hospital, clinic, HMO or federal government agency. 2 These 
eligibility criteria are very similar to those used for the SMS. Data on practice expenses are 
for calendar year 1988. 

The PPCIS contains 3,505 observations, similar to the size of the SMS. However, the 
419 physician employees in the sample have been excluded from this analysis because they 



2 A detailed discussion of eligibility criteria and the survey can be found in Dayhoff etal . (1992). 



gpci\ pracexpX rent\ chap5 



II-5-3 



were not asked financial questions about their practices. In addition, some physicians chose 
not to respond to the questions regarding office expenses. For this analysis, we do not utilize 
imputed data, leaving a sample of 2,265 observations. 

The office expense section of the PPCIS contains a series of questions. Physicians 
were first asked whether they rented or owned their office space. Renters were asked to 
report their yearly rental or lease cost and owners to report their total yearly tax deductible 
depreciation and interest expenses. If this amount did not include utilities and telephone, 
physicians were asked to provide these expenses separately. By summing the responses to 
these three questions, we obtain the annual expenditures for office space; including utilities 
and telephone. 

Physicians were asked to provide total costs for the group in which they practiced, 
unlike the SMS which asks for the physician's share of costs. Thus, cost per square foot of 
office space can be calculated by dividing the practice's total expenses for rent, taxes and 
depreciation, and utilities by the total square footage reported on the survey. This eliminates 
the need to assume that each physician owner shares equally in expenses, as must be done 
using the SMS data. 

The PPCIS questions are quite detailed, attempting to prompt physicians to ensure 
that all relevant costs are provided, and verifying the initial response with physicians who 
reported no expenses for office space costs. However, these data are still not ideal for 
measuring physicians' true office space costs. First, all expense data on the PPCIS are self- 
reported; no audit is conducted to verify that amounts reported by respondents are accurate 
values. (This is also true of the SMS.) Second, while the PPCIS asks for details on the office 
arrangements of physicians reporting no office expenses, it obtains much less detail on 
physicians who do have expenses. For instance, the PPCIS identifies physicians with no 
office expenses because they received free space from a hospital; it does not, however, 
identify physicians who received subsidized space from a hospital. 

5.2.2 Descriptive Results 

Given a survey containing data for 2,265 physicians nationally, sample sizes for small 
geographic areas (i.e., counties, MSAs) are too limited to provide reliable estimates of office 
space costs. However, for broader geographic areas, data from the PPCIS can be 
meaningfully compared with GPCI rental indices computed from HUD data to verify the 
accuracy of using apartment rent data as a proxy for physician office space. PPCIS 



gpci\ pracexp\ rent\ chap5 



II-5-4 



observations were indexed by dividing the median for each geographic area by the sample 
median. For comparison, we use 1993 HUD fair market rents, with the value for New York 
City replaced by that for Bergen Passaic, indexed to a population-weighted national mean of 
1.0. 

Table 5-1 presents a comparison of PPCIS and GPCI index values by urbanicity, 
Census region, Census Division, and for selected (,;fia s with relatively large numbers of 
observations. The PPCIS indicates urbanicity of practice using three categories: rural, small 
MSA (under one million population) and large MSA (over one million population). The 
PPCIS and the GPCI yield very similar index values for both urban and rural locations. The 
largest difference between PPCIS and GPCI values is for small MSAs (1.01 compared to .92); 
for other stratifications the indexes differ by only a few percentage points. 

The differences between PPCIS and GPCI indexes for the four Census regions are 
noticeably larger. The GPCI for the Northeast is about 15 percentage points higher than the 
PPCIS index; the GPCI for the West is about 7 percentage points higher than its PPCIS 
counterpart. Conversely, GPCIs for the Midwest and South are lower than the PPCIS 
indexes. Indexes by Census Division provide a similar pattern; GPCI values are greater than 
PPCIS values for the New England, Middle Atlantic, and Pacific Divisions. 

The reasons for these differences in index values are not obvious. One possibility is 
that the relationship between rents for residential apartments and for physician office space 
are systematically different across regions. Under this interpretation, the GPCI values in the 
Northeast and West overstate the true costs of physician offices and those in the South and 
Midwest understate the true costs. However, this is not the only possible explanation for the 
differences. From conversations with real estate brokers (see Chapter 2) we know that 
physician offices typically require significant improvements (electrical, lighting, plumbing) 
that raise the cost of office space, relative to other commercial space. In Part III of this 
report, it is found using the PPCIS data that physicians in the Northeast have much lower 
expenses for medical supplies, medical equipment, and miscellaneous expenses (from 15 to 
30 percent) than those practicing in other regions. Our analysis of the PPCIS indicates that 
physicians in the Northeast utilize smaller offices than do physicians in other regions. This 
suggests physicians in the Northeast may do fewer tests and procedures in their offices than 
those located in other regions. If this were the case they would require fewer such 
improvements, thereby reducing their costs. Thus, variations in expensiveness of 
construction, site-of-service (office/ nonoffice), type of office space, practice style (and 



gpci\pracexp\rent\chap5 



11-5-5 



TABLE 5-1 



COMPARISON OF PHYSICIAN OFFICE EXPENSE PER SQUARE FOOT TO FAIR MARKET 
APARTMENT RENTS (Indexed to a National Average of 1.0) 



EXPENSE PER 
SQUARE FOOT 



N 



Index 1 



FAIR MARKET RENT 



Index 2 



URBANICITY 

Rural 
Urban 



544 
1,721 



0.75 
1.07 



0.74 
1.08 



Small MSA 
Large MSA 



1,066 
655 



1.01 
1.19 



0.92 
1.19 



REGION 

Northeast 
Midwest 
South 
West 



447 
490 
802 
526 



1.07 
0.95 
0.91 
1.13 



1.22 
0.87 
0.84 
1.19 



DIVISION 

New England 
Middle Atlantic 
East North Central 
West North Central 
South Atlantic 
East South Central 
West South Central 
Mountain 
Pacific 



132 
314 
327 
163 
418 
143 
242 
136 
390 



1.15 
1.05 
0.95 
0.97 
0.93 
0.81 
0.94 
0.98 
1.18 



1.22 
1.26 
0.89 
0.81 
0.92 
0.70 
0.79 
0.91 
1.29 



MSA 

Chicago 
Los Angeles 
Oakland 
San Francisco 



40 
77 
22 
24 



1.27 
1.33 
1.19 
1.54 



1.22 
1.45 
1.46 
1.75 



'Indexed to a national median of 1 .0. 

indexed to a population-weighted national average of 1 .0. 

SOURCE: 1988 Physician's Practice Costs and Income Survey and FY 1993 HUD Fair Market Rents. 



II-5-6 



gpci\pracexp\rent\TAB5-1 XLS 



specialty) across regions could account for differences in actual expenditures per square foot 
relative to GPCI values. 

The bottom panel of Table 5-1 presents PPCIS and GPCI values for four metropolitan 
areas. Sample sizes (N) indicate the problem with performing these type of comparisons for 
small geographic areas; even the most populous MSAs often have fewer than 40 practices 
providing data for office expenses on the PPCIS. Nevertheless, it is interesting to note the 
general correspondence between relative physician office expense per square foot and 
apartment rents, although the apartment rent index is higher in three of four cities. 

5.2.3 Regression Results 

To further explore the relationship between the GPCI and reported physician 
expenditures, we perform regression analysis with the PPCIS similar to that done previously 
by Gillis et al. and Zuckerman et al. using the SMS data. Following their methodology, we 
begin by regressing the natural log of cost per square foot from the GPCI on a series of 
locality dummy variables. The resulting F-statistic on the set of dummy variables tests the 
hypothesis that there is no significant variation in the input price by locality. As expected, 
this hypothesis is rejected; the F-statistic of 2.04 implies less than one in one thousand chance 
that the costs do not vary by locality. 

To estimate the correlation between the PPCIS cost data and the GPCI values, we 
regress the natural log of cost per square foot on the natural log of the GPCI, as shown in 
equation 2. The proper specification for X, the vector of physician characteristics is unclear. 
It may be appropriate to include specialty dummy variables since office space requirements 
may vary systematically by specialty. However, as Gillis et al. note, the distribution of 
physicians by specialty differs systematically across areas, with primary care physicians being 
more prevalent in rural areas. Given differences in the GPCIs across rural and urban areas, it 
is possible that specialty dummies will pick up input price variation rather than control for 
differences in input characteristics. Thus, we performed regressions using alternative 
specifications with and without specialty dummy variables. 

Our other control variable, OFFTYPE, reflects whether the physician rents or owns 
office space. The costs of physicians who own space may be more representative of previous 
market conditions (at the time of purchase) rather than current costs, given multi-year 
mortgages with fixed annual payments. Since we know little about the geographic 



gpci\ pracexp\ rent\ cha p5 



II-5-7 



distribution of owners or the age of their mortgages, we include this dummy variable in 
some of our regression specifications. 

The results of these regressions are presented in Table 5-2. Each of the alternative 
models utilizes the natural log of cost per square foot from the PPCIS as the dependent 
variable. Models 1, 3, and 5 use the natural log of a the office rental GPCI index merged 
onto the PPCIS by the county as an independent variable. The GPCI index is based on 1993 
HUD FMRs. Models 2, 4, and 6 replaces the county level GPCI index with the GPCI index 
constructed for localities. Models 1 and 2 include both specialty and office type dummy 
variables; 3 and 4 contain office type dummy variables; 5 and 6 include ln(GPCI) as the only 
explanatory variable. 

The coefficients on ln(GPCI) are very similar across all six specifications, ranging 
from 0.62 to 0.70. All are significantly greater than zero, implying that the GPCI rental index 
is correlated with office costs reported on the PPCIS. However, all the coefficients are 
significantly less than one, implying the relationship is less than proportional. That is, these 
estimates imply that a 10% increase in the GPCI corresponds to roughly a 7% increase in 
physician office expense per square foot. These elasticity estimates are very similar to those 
found by Gillis et ah and Zuckerman et ah in their work with the SMS. 3 

To investigate why these elasticities differ from one, we re-estimated model 5, using 
the county level GPCI, for several subsamples from the PPCIS. We began by estimating the 
model separately for the three office types: physicians who rent, physicians who own, and 
physicians who both rent and own. The coefficients from thes^ regressions explain the 
elasticity between PPCIS and GPCI data within each particular subsample. For example, we 
had speculated that the coefficient from a regression including only physicians who own 
would be relatively low, since their costs reflect market conditions from the time t 
mortgage was signed, rather than current conditions. 

The elasticity estimates from these regressions are presented in Table 5-3. As 
expected, the coefficient for physicians who rent (0.70) is greater than that for physicians who 
own office space (0.63). However, both are significantly lower than one. The estimate for 
physicians who both rent and own is much lower, however, relatively few physicians have 
this practice arrangement. 



The time lag between the 1993 HUM FMRs and the 1988 PPCIS appears to have little effect on the results. For comparison, 
we estimated the regressions using the GPCI constructed with 1987 HUD data, and the results were virtually identical. 



gpci\pracexp\rent\chap5 



II-5-8 



TABLE 5-2 



REGRESSIONS OF PPCIS COST PER SQUARE FOOT ON GPCI OFFICE RENTAL INDEX 





Model 1 


Model 2 


Model 3 


Model A 


Price Proxy 










Log County Based GPCI 


0.63 




0.67 






(12.01) 




(13.11) 




Log Locality Based GPCI 




0.62 




C.67 






(11.00) 




(12.00) 


Office Type a 










Physician Rents 


-0.19 


-0.19 


-0.19 


-0.19 




(3.22) 


(3.16) 


(3.25) 


(3.17) 


Physician Owns 


-0.74 


-0.74 


-0.77 


-0.77 




(10.56) 


(10.52) 


(10.95) 


(10.93) 


Specialty 










GP/FP 


-0 31 


-0 33 

U. J J 








(3.36) 


f3 62) 






Internal Medicine 

III V v 1 1 I l>4 1 1 W 1 W VJ 1 VI 1 1 W 


-0.18 


-0.19 








(i 98) 


(2 02) 






Cardiology 


-0.25 


-0.24 








11 29) 


C2 20) 






Gastroenterology 


-0.16 


-0.15 








(1 47) 


n 43) 






Other Medical 


-0.16 


-0.15 








(1 64) 


(1 67) 






General Surgery 


-0.12 


-0.13 








(1 26) 


(1 36) 






Orthopedic Surgery 


-0.01 


-0.19 








(0 07) 


(0 18) 






Ophthalmology 


-0.67 


-0.08 








-(0.66) 


(0 75) 






Urology 


-0.10 


-0.11 










M 10) 






Ob/Gvn 


-0 14 


-0 15 








M 44} 








Thoracic Suraerv 


03 

W . WW 


03 

W. WW 








(0.30) 


(0.31) 






Other Surgery 


-0.05 


-0.05 








(0 55) 


(0 54) 






Psychiatry 


-0.02 


-0.02 








(0.25) 


(0.24) 






Anesthesiology 


-0.29 


-0.29 








(2.46) 


(2.48) 






Radiology 


0.26 


-0.24 








(0.22) 


(0.21) 






Intercept 


3.07 


3.07 


2.95 


2.94 


R-Square 


0.16 


0.16 


0.15 


0.14 



Model a 

0.70 

(13.23) 



Model 6 



0.70 
(12.07) 



2.69 
0.07 



2.69 
0.06 



a 

Physician both rents and owns is the omitted category. 
b Other specialties (primarily pathology and emergency medicine) is the omitted category. 

NOTE: t-statistics are in parentheses. 

SOURCE: 1988 Physician's Practice Costs and Income Survey and HUD Fair Market Rents. 



gpci\pracexp\rent\TAB5-2.XLS 
II-5-9 



TABLE 5-3 



ELASTICITY ESTIMATES FOR SUBGROUPS FROM REGRESSION OF PPCIS EXPENSE 
PER SQUARE FOOT ON GPCI INDEX OFFICE RENTAL 



Subgroup 



Elasticity 
Estimate 



t-statistic 



Standard 
Error 



OFFICE TYPE 

Physician Rents 
Physician Owns 
Physician Both Rents & Owns 



0.70 
0.63 
0.33 



13.39 
3.27 
1.58 



0.05 
0.19 
0.21 



URBANICITY 

Rural 0.46 2.96 0.16 

Small MSA 0.78 6.20 0.13 

Large MSA 0.44 3.43 0.13 



REGION 

Northeast 0.77 5.36 0.14 

Midwest 1.11 5.69 0.19 

South 0.70 6.96 0.10 

West 0.61 4.97 0.12 



Source: 1988 Physicians' Practice Costs and Income Survey and 1993 HUD Fair Market Rents. 



II-5-10 



gpci\pracexp\rent\TAB5-3.XLS 



Regressions run separately for three levels of urbanicity indicate that the GPCIs are 
more proportional with PPCIS data for small MSAs (less than one million population) than 
for rural areas or large MSAs. For small MSAs we cannot reject the hypothesis that the 
elasticity is equal to one; for other areas the estimate is clearly less than one. 

Our final stratification provides estimates separately by Census region. Here we find 
that the estimates for the Northeast and Midwest are not statistically different from one. 
While the elasticities for the South and West are statistically greater than zero, they are also 
statistically less than one. The reasons for differing elasticities among these subsamples 
stratified by location are not clear. Sample sizes from the PPCIS prevent further 
stratifications—for example by urbanicity and region-that might help determine whether it is 
the region of the country, or the prevalence of each urbanicity type among regions that is 
driving these results. 

The finding of elasticity estimates less than one using data from both the SMS and the 
PPCIS implies that physician expenditures per square foot of office space are less variable 
across areas than are HUD FMRs. One possible explanation for this result, the presence of 
physicians who own office space in physician surveys was tested using PPCIS data, and was 
found to have little effect. Two alternative explanations cannot be tested using our data. 
First, it is possible that physicians in high rent areas choose to locate in less-expensive types 
of offices, perhaps because they are more likely to do tests and procedures at non-office sites 
such as hospitals. (High rent areas tend to be large cities with a greater prevalence of non- 
office sites.) This would compress the distribution of physician expenditures per square foot 
relative to the distribution of apartment rents. Second, it is possible that random error in the 
FMRs' measurement of apartment rents is biasing the regression elasticity towards zero. 
Although the FMRs are generally accurate measures of apartment rents, for particula r areas 
in particular years they may over- or under-state actual rents. The existence of random error 
in the FMRs is attested to by the sometimes large changes in FMRs that occur when area 
rents are resurveyed (see Chapter 3). In addition, the FMRs do not measure rental variations 
within metropolitan areas. With random error in the FMRs, even if the true relationship 
between apartment rents and physician office rents is proportional, the measured elasticity 
would be less than one. 



gpci\ pracexp\ rent\ chap5 



II-5-11 



5.2.4 Variation in Costs After Adjusting for the GPCJ 

Gillis et al. investigated whether systematic differences in input prices remain after 
adjusting the cost per square foot data from the SMS by the GPCI, and found evidence of 
significant variation in GPCI-adjusted prices across localities. We replicated this work using 
PPCIS data. The PPCIS cost per square foot was deflated by the GPCI, and the natural log of 
this ratio was regressed on a series of locality dummy variables. The joint F-test for 
significance on all the locality variables indicated that the hypothesis that GPCI-adjusted 
prices were the same could be rejected. (Table 5-1) 

To characterize areas where the GPCI-adjusted prices appeared to be systematically 
high or low, we regressed the natural log of the GPCI-adjusted values on three dummy 
variables for varying levels of urbanicity --rural, small MSAs, and large MSAs. The middle 
panel of Table 5-4 reports the results of this regression. The joint F-test indicates that the 
adjusted prices do vary significantly across these three levels of urbanicity. The positive and 
significant coefficient for small MSAs implies that the GPCI-adjusted values for these areas 
are larger than those found for large MSAs (the omitted category). This is consistent with 
our descriptive findings reported in Table 5-1 in which the difference in median mean values 
between the PPCIS and GPCI was greater for small MSAs than for large MSAs or rural areas. 
The bottom panel presents results of a similar regression with dummy variables for region 
replacing the urbanicity variables. The GPCI adjusted costs do vary by region, with the 
South and Midwest significantly higher than the Northeast. 

5.2.5 Conclusion 

This analysis compares the HUD FMR data with actual physician expenditures per 
square foot of office space reported on the ?PCIS. We find that the FMRs are positively 
correlated with office expenses. However, estimates of the elasticity of expenses per square 
foot with respect to the GPCI rental index are less than one— indicating that the GPCI exhibits 
greater variation across areas than does office expenses as reported on the PPCIS. These 
results are similar to those found in previous studies utilizing physician expenditure data 
from the SMS. However, these results do not necessarily indicate a fundamental problem 
with the use of HUD FMR data as a proxy for the price of physician office space. Differences 
in the type of office space (associated with geographic differences in site of service), random 
error in the FMRs, and geographic differences in the prevalence of subsidized space from 



gpci\ pracexp\ rent\ chap5 



II-5-12 



TABLE 5-4 



TESTS FOR VARIATION IN GPCI ADJUSTED EXPENSE PER SQUARE FOOT 



Log (COST/GPCn 



Joint F-statistic 



R-square 



Carrier-Localities 

(significance level) 



1.65 
0.0001 



0.22 



Coefficient 



t-statistic 



Urbanicity Dummies 
Rural 

Small MSA 



-0.039 
0.165 



0.93 
4.62 



Coefficient 



t-statistic 



Region Dummies 

Midwest 

South 

West 



0.138 
0.161 
0.077 



2.90 
3.76 
1.64 



SOURCE: 1988 Physicians' Practice Costs and Income Survey and 1993 HUD Fair Market Rents. 



II-5-13 



gpci\pracexp\rent\TAB5-4.XLS 



hospitals are all factors that could help account for a non-proportional relationship between 
the GPCI index and physician office expenditures. We also find that variation remains in 
PPCIS cost per square foot data deflated by the GPCI. However, divergences between the 
GPCI and physician expenditure data do not indicate that the GPCI is systematically low in 
rural areas or high in urban areas. Instead, we find the greatest differences occur across 
regions— in particular, low office expenses in the Northeast relative to apartment rents. 
Previous analysis based on the PPCIS suggest that this stems from a practice style in the 
Northeast that is less office-based than found in other regions. 



gpci \ pracexp\ rent\ cha p5 



II-5-14 



6.0 



FINAL OFFICE RENTAL INDICES 



After reviewing the earlier analyses in this report, the Health Care Financing 
Administration (HCFA) made the following decisions: 

• the Fair Market Rents (FMRs) established by the Department of Housing and 
Urban Development (HUD) should continue to be used to define the GPCI 
office rental index; 

• the GPCI office rental index should be based on 2-bedroom FMRs rather than 
4-bedroom FMRs; 

• the single most recent year of FMRs available should be used for the GPCI 
office rental index. For the 1996 GPCI update, the most recent available FMRs 
are the FY 1994 FMRs; 

• county-specific rents (rather than metropolitan wide rents) should be reflected 
within large metropolitan areas (CMSAs), using the HUD special tabulation of 
1990 Census apartment rents; and 

• the actual New York City FMR should be used in the GPCI rental index. 

In addition, HCFA decided that the GPCI should be weighted by relative value units 
or RVUs (practice expense RVUs in the case of the office rental index) rather than by 
population. Weighting by RVUs has two advantages. First, county RVUs identify the 
counties where Medicare services are provided more accurately than county population. The 
GPCI should be weighted to reflect input prices in the locations where services are 
performed. Second, weighting by RVUs assures that aggregate Medicare Fee Schedule 
payments for practice expense will not change if the payment locality configuration is altered 
(e.g., if a state with multiple localities becomes a single statewide locality). The differences 
between the RVU-weighted office rental index and the population-weighted office rental 
index are shown in Appendix II-2, where payment localities are ranked in descending order 
of the difference between the two indices. 1 The change in the office rental index from RVU- 
weighting ranges from a 8.1 percent increase in South Dakota to a 11.7 percent decline in the 
Rockford, Illinois locality. The changes in most localities are much smaller. 

The 1996 updated office rent index differs from the 1992 office rent index in the 
following respects: 



The weight affects the computation of the wage index in two places. The first is in normalizing county rents by national 
average rents. For the final office rental index, the national average is weighted by practice expense RVUs rather than by 
population. This has the effect of scaling each payment locality's index up or down by the same proportion. The second effect 
occurs in taking a weighted average of county or county part values to form an index value for a payment locality. For the final 
office rental index, this average is weighted by practice expense RVUs rather than by population. This change affects each 
locality's index according to the difference in distribution among counties/ county parts of RVUs versus population, and the 
difference among counties/ county parts of rents. 



gpci\ pracexp\ rent\ chap6 



II-6-1 



based on final FY 1994 FMRs rather than FY 1987 FMRs. The FY 1994 FMRs 
reflect a "benchmark" revision based on the 1990 Census, and also incorporate 
post-1990-Census rental survey data; 

• based on 2-bedroom rather than 4-bedroom FMRs; 

• reflects OMB's June 1993 redefinitions of metropolitan areas; 2 

• RVU-weighted rather than population-weighted; 

• county-specific rents, rather than metropolitan area rents, are measured within 
large metropolitan areas (CMSAs); and 

• New York City's actual FMR is used, rather than being replaced by the FMR of 
Bergen-Passaic New Jersey. 

Of these changes, the use of the more recent FY 1994 FMRs has by far the largest overall 
effect on the office rental index. 

After the original computation of the 1996 practice expense GPCI, HCFA actuaries 
determined that it would have to be rescaled to achieve "budget neutrality", that is, to ensure 
that aggregate payments to physicians are not affected by updating the GPCI. The practice 
expense GPCI was therefore multiplied by the factor 1.00125 and rounded to three decimal 
places. Since the practice expense GPCI is a linear combination of the employee wage and 
office rental indices (plus a constant representing the share of supplies, equipment, and 
miscellaneous in practice expenses), rescaling the practice expense GPCI in effect rescales the 
office rental index by the same factor. In Appendix II-3, we show the 1996 office rental 
index, the rescaled 1996 office rental index (the 1996 index multiplied by 1.00125), and the 
1992 office rental index, by Medicare payment locality. 

Appendix IT4 ranks Medicare payment localities by descending order of difference 
between the 1996 rescaled office rental index and the 1992 office rental index. The changes 
from updating are significant. Twenty of the 216 Medicare payment localities gain more than 
10 percentage points and 55 lose more than 10 percentage points. The largest gair. r is 
Hawaii, whose index rises 52 percentage points, or 41 percent. The largest loser is Victoria, 
Texas, which loses 38 percentage points. Nevertheless, it should be remembered that the 
office rental index accounts for only approximately 10 percent of the overall Medicare Fee 
Schedule Geographic Adjustment Factor (GAF). Hence, even the largest changes in the office 
rental index imply a gain or loss of less than 5 percent in the total Medicare fee. 



2 Although the office rental index is ultimately calculated for Medicare payment localities, the underlying rental data is 
tabulated by HUD for metropolitan areas and nonmetropolitan counties before it is crosswalked to localities. 



gpci\pracexp\rent\chap6 



II-6-2 



REFERENCES 



Building Owners and Managers Association International, 1992 BOMA Experience Exchange 
Report, Washington, DC, 1992. 

College of Physicians & Surgeons of Columbia University, letter to Gail Wilensky at HCFA, 
September 20, 1991. 

Dayhoff, D. and G. Pope, Validation of Medicare Fee Schedule Office Rental Proxy, Final 
Report, PPRC Contract No. T74276093/1, July 1990. 

Dayhoff, D., et al „ Handbook for Using the 1988 Physician's Practice Costs and Income 
Survey, Final Report, HCFA Cooperative Agreement No. 99-C-98526/1-08, April 1992. 

DeGroot, M. H., Probability and Statistics . Reading, MA: Addison-Wesley Publishing Co., 
1975. 

Gillis, K, R. Reynolds, and R. Willke, "Assessing the Validity of the Geographic Practice Cost 
Indexes", Chicago: American Medical Association Center for Health Policy Research, 
September 1992. 

Gonzalez, M., ed., Socioeconomic Characteristics of Medical Practice, 1992, Chicago: 
American Medical Association Center for Health Policy Research, 1992. 

Institute of Real Estate Management, 1992 Office Buildings: Income/ Expense Analysis, 
Chicago: IREM of the National Association of Realtors, 1992. 

Medical Society of the State of New York, letrjr to Gail Wilensky at HCFA from Charles 
Aswad, MD, August 5, 1991. 

Physician Payment Review Commission, Annual Report to Congress 1992, Washington, DC, 
1992. 

Schmitz, R. J., Estimating Physician Practice Costs in Manhattan, report to the College of 
Physicians & Surgeons of Columbia University, September 1991. 

Society of Industrial and Office Realtors, 1992 Guide- Comparative Statistics of Industrial and 
Office Real Estate Markets, Washington, DC: SIOR and Landauer Real Estate Counselors, 
1992. 

U.S. Department of Housing and Uban Development, Report to Congress on Rent Control, 
Washington, DC: U.S. Government Printing Office, September 1991. 

Urban Land Institute, Market Profiles 1992, Washington DC, 1992. 

Welch, W. P., S. Zuckerman, and G. Pope, The Geographic Medicare Economic Index: 
Alternative Approaches, HCFA Grant No. 18-C-98326/1-01, June 1989. 



gpci\ pracexp\ rent\ refs 



II-R-1 



Welch, W.P., S. Zuckerman, and G.C. Pope, Supplement to the Geographic Medicare 
Economic Index: Alternative Approaches, revised August, 1990. HCFA Grant No. 17-C- 
99222, 18-C-98526, and 17-C-98758. NTIS No. PB91-113506. 

Zuckerman, S., M. Miller, M. Wade, and M. Pauly., Regional Variation in the Impact of 
Medicare Physician Payment Reform, Final Report, HCFA Contract No. 500-89-0054/3, 
November 1992, NTIS No. PB94-104056. 

Zuckerman, S., P. Welch, and G. Pope, The Development of an Interim Geographic Medicare 
Economic Index, HCFA Grant No. 18-C-98326/1-01, December 1987. 



gpci\ pracexp\ rent\ refs 



II-R-2 



part nt 

SUPPLIES, EQUIPMENT, AND 
MISCELLANEOUS PRICES 



PART IIL SUPPLIES, EQUIPMENT, AND MISCELLANEOUS PRICES 



1.0 INTRODUCnON 

The Geographic Practice Cost Index (GPCI) is used in Medicare's Fee Schedule for 
physician services to adjust payments for geographic variation in physicians' practice costs. 
Three components of the practice expense GPCI are medical supplies, medical equipment, 
and other (miscellaneous) expenses. Together they comprise approximately one-third of the 
practice expense GPCI. Currently, there is no adjustment for geographic variation in the 
prices of these three inputs. It has been assumed that physicians could purchase supplies, 
equipment, and miscellaneous inputs in "national markets" at the same price everywhere 
(Welch, Zuckerman, and Pope, 1989). However, no data has been located to verify this 
assumption. 

1.1 Overview of Part IH of the Report 

Part III of this report reevaluates geographic variation in the prices of physician 
supplies, equipment, and miscellaneous inputs. Chapter 2 of Part III considers the 
availability of price data by area for these inputs, and also describes anecdotal/ interview 
evidence on price variation. First, we review several previous searches for such data, and 
then describe the results of our additional efforts. Our conclusion is that price data are not 
currently available. Therefore, we then discuss issues in collection of primary data on input 
prices. In Chapter 3, we analyze geographic variation in expense and equipment purchase 
price data from HCFA's 1988 Physicians' Practice Cost and Income Survey. This analysis is 
contained in Section 3.2. In Section 3.1, we review two previous studies that analyzed 
geographic variation in AMA physician survey expense data and also discuss the limitations 
of inferring geographic variation in prices from geographic variation in expenditures . 
Chapter 4 provides conclusions and discusses the Health Care Financing Administration's 
decision not to make an adjustment for supplies, equipment, and miscellaneous prices in the 
updated (1996) practice expense GPCI. 



gpci\pracexp\sem\chapl 



III-l-l 



2.0 INPUT PRICE DATA 



The subject of this chapter is the nonlabor inputs physicians use, what is known about 
the prices of these inputs, and the how better price data by area might be obtained. This 
chapter thus focuses on medical supplies, medical equipment, and miscellaneous inputs. 
These three inputs comprise about one third of the practice expense GPCI. 

Medical supplies consist of a wide range of items consumed in physician offices. 
Among the major subcategories of medical supplies are pharmaceuticals, dressings, catheters, 
syringes, tubes, and X-ray film. Within the pharmaceutical category alone, there are literally 
hundreds, if not thousands, of drugs and other ethical preparations used in performing 
physician services. IMS America, a major marketing research firm, classifies non- 
pharmaceutical medical supplies into 38 major categories. Each of these major categories 
usually consists of several sub-categories of products. As with pharmaceuticals, there are 
hundreds, if not thousands, of medical supplies. 

There is also a wide variety of medical equipment used by physicians in their offices. 
Prominent among them are diagnostic equipment (e.g., X-ray machines and ultrasound 
machines) and laboratory equipment such as clinical chemistry analysis machines. Unlike 
medical supplies, which tend to be used just once, medical equipment has a physical 
economic lifespan of several years. Equipment such as CT-scanners that used to be 
associated only with hospitals is now starting to appear in physician offices. While the 
variety of medical equipment is high, it is not as great as medical supplies. 

The miscellaneous category is quite heterogeneous. Professional automobile, 
continuing education, legal services, accounting services, office management services, 
professional association memberships, journals, telephone services, and office furniture are 
among the items that constitute this category. The most recent survey of physician practice 
expenses conducted by Medical Economics (1992) indicates that professional automobile 
expenses comprise 1.4 percent of total practice receipts and 3.9 percent of non-physician 
expenses while continuing education comprises 0.9 percent of total practice receipts and 2.7 
percent of non-physician expenses. 

2.1 Secondary Data 

In this section we review the findings of previous studies that tried to obtain 
secondary input price data and then present the findings of our new search efforts. 



gpci\ pracexp\ sem\ chap2 



III-2-1 



2.1.1 Review of Previous Searches for Secondary Data 

Urban Institute Report to HCFA 

As part of a study on regional variation in the impact of recent Medicare physician 
payment changes, the Urban Institute (1992) assessed the practice expense GPCI. The Urban 
Institute tried to ascertain the degree of geographic input price variation in medical 
equipment by contacting trade associations and manufacturers of medical equipment. The 
National Association of Medical Equipment (NAME), "representing home health and 
rehabilitation equipment suppliers, claimed that they did not track prices in different 
geographic areas." The Health Industry Manufacturers Association (HIMA), representing 
major producers of medical equipment, had done research on geographic variation in 
equipment prices. However, citing proprietary data, HIMA refused to relate to the Urban 
Institute their findings. The inability and/ or refusal by these associations to cooperate with 
the Urban Institute is particularly disappointing because these associations have the potential 
to provide geographic price information on a large variety of equipment types rather than 
just one or two items. The Urban Institute also investigated the Current Industrial Reports 
series produced by the U.S. Bureau of the Census. The Census data, which is time series in 
nature, is not suited for studying geographic price variation. 

Based on information from the Medical Device Registry and the Health Industry 
Buyers Guide, the Urban Institute identified and contacted four major manufacturers of 
medical equipment. The manufacturers indicated that there is some variation in the prices 
physicians pay for medical equipment with volume discounts accounting for a significant 
share of the price variation. However, the manufacturers did not indicate to the Urban 
Institute that the physician's geographic location was a factor in equipment price variation. 
Finally, following a suggestion from the Health Industry Distributive Association (HIDA), the 
Urban Institute contacted IMS America, a firm which provides marketing services to medical 
supply companies. Since the Center for Health Economics Research (CHER) had also 
contacted IMS for a study of hospital nonlabor input prices, discussion of IMS data is 
deferred to a review of our findings. Aside from (perhaps) IMS, the Urban Institute 
concluded that there were no reliable sources of input price data for medical supplies, 
medical equipment, and miscellaneous expenses which could be used to geographically 
adjust physician payments. 



gpd\pracexp\sem\chap2 



III-2-2 



CHER Hospital Nonlabor Index Report to HCFA 

In a study of the geographic variation of nonlabor input prices paid by hospitals, 
CHER (1992) obtained and appraised data collected by both public and private organizations. 
Since CHER's report was an update and extension of a report by ProPAC (1989) on the same 
subject, the review encompasses ProPAC's results. 

CHER obtained seven secondary input price data series - about the same inputs as the 
ProPAC obtained. Only one of the data series CHER obtained, commercial telephone, might 
be part of the physician practice market basket. The other items either are not part of market 
basket (e.g., food) or are examined in another GPCI report (e.g., malpractice insurance). The 
prices of all of the hospital nonlabor inputs exhibited geographic variation. However, both 
CHER and ProPAC found that while all of the input prices varied geographically, the prices 
were not highly correlated, and some correlations were even negative. Thus, if a nonlabor 
input price index were constructed, it would exhibit less geographic variability than most of 
its constituents parts. 

IMS America collects price data on medical supplies and pharmaceuticals purchased 
by 350 hospitals. According to IMS, their sample is a statistically representative sample of 
seven hospital size groups and the nine census divisions. 1 IMS collects paid invoices from 
the participating hospitals on a monthly basis. The invoices may reflect discounts from the 
manufacturer or distributor if they are given on specific products and not on the tctal 
amount purchased. Any discounts offered after the time at which the invoice was issued, 
such as rebates, are not recorded. The inconsistent handling oi price discounts by IMS raises 
the possibility that its data might not be of sufficient quality for use in a geographic price 
index. 

The IMS data is proprietary. IMS produces, for a substantial fee, both standardized 
and customized reports for its customers. Thus the quality of the IMS data has not been 
independently assessed. Since IMS has yet to disclose the identity of its participating 
hospitals, its claim that its sample is representative also cannot be verified at this time. Even 
if the sample is representative with regard to its primary sampling criteria, the sample may 
not be random. Many hospitals refuse to participate, even though IMS provides participating 
hospitals with free market share reports. Further, the sample size may be too small. Without 
more cooperation from IMS and substantial resources to assess the quality of the IMS data, 
IMS remains a source which may or may not prove to be useful in constructing price indices 



'Secondary selection criteria include urbanicity, teaching status, and purchasing group participation. 
gpd\pracexp\sem\chap2 II 1-2-3 



for reimbursing hospitals and/ or physicians. Finally, even if the IMS data were to prove 
adequate for hospital reimbursement, physicians might use medical supplies that hospitals do 
not purchase. 

2.1.2 Results of HER Search Efforts 

Rather than duplicate the Urban Institute's and CHER's previous efforts in contacting 
national trade associations, medical suppliers and equipment manufacturers, and IMS in 
order to determine the extent of geographic pric ?. variation, we decided to pursue new 
avenues of information. To this end, we contacted wholesalers and local distributors to gain 
insights how prices might vary geographically. Transportation firms were also contacted 
since the prices physicians pay for nonlabor input presumably reflect shipping costs. 

Medical Supplies and Medical Equipment 

In addition to shipping costs (discussed in the following section), there are several 
potential scurces of geographic price variation. First, manufacturers might practice 
geographical price discrimination. That is, aside from shipping costs and volume discounts, 
manufacturers may charge different prices, for the same item, to distributors in different 
parts of the country. 2 Second, distributors' costs of doing business might vary 
geographically. And third, the degree of competition between local distributors might also 
affect the markup charged to physicians. 

The first thing we did was to establish the degree to which physicians purchased 
medical supplies and equipment from local distributors, regional/ national distributors, and 
directly from manufacturers. This was done to ascertain to what extent the market for these 
items are national or local/ regional. For the most part, manufacturers do not directly process 
orders from individual physicians and physician groups unless the order is for medical 
equipment in excess of 20-25 thousand dollars. As noted earlier, manufacturers did not 
indicate to the Urban Institute that physician location, aside from shipping costs, played a 
role in equipment prices. This suggests that the market for X-ray machines, CT-scans, and 
MRIs is national. Of these expensive items, however, only X-ray machines are commonly 
present in physician offices. 



Arbitrage (elimination of price differentials) may be too costly to eliminate geographic price differentials. 
gpci\pracexp\sem\chap2 III-2-4 



As was found in previous efforts to obtain information on whether prices vary 
geographically, there is a dearth of hard data. There is some anecdotal evidence that some 
manufacturers charge west coast distributors more than in other areas. Several distributor 
and manufacturer representatives thought that the reason why west coast distributors would 
pay higher prices (instead of shipping in cheaper supplies/ equipment from other areas) is 
because there are fewer distributors on the west coast and, therefore, less intense price 
competition between distributors. However, Baxter Healthcare, a major manufacturer of 
medical supplies, stated that except for volume discounts, it sold its products to all 
distributors at the same price. Additionally, ABCO Dealers, Inc., a cooperative of distributors 
located over the entire country, stated that, aside from shipping costs and volume discounts, 
distributors paid the same price regardless of location. 

AmeriNet, a non-profit buying group operating out of St. Louis, sells a large range of 
products including pharmaceuticals, medical supplies, and medical equipment to customers 
over the entire country. AmeriNet sells mostly to hospitals, health maintenance 
organizations, nursing homes, and physician groups. Except for transportation costs, the 
prices AmeriNet charges do not vary geographically. Because of low volumes, AmeriNet 
does not sell products to individual physicians. Individual physicians tend to buy from local 
distributors. Because of local differences in the cost of business, prices charged to individual 
physicians might geographically vary as much as 20 percent (an estimate supplied by 
AmeriNet). 

With regard to urban/ rural differences, one distributor organization indicated that 
rural physicians pay higher prices than urban physicians. The reason offered is that rural 
practices are visited less by sales representatives of distributors than urban practices. Thus, 
the degree of competition among distributors may affect input prices paid by physicians. A 
distributor in the northeast, however, stated that rural physicians paid the same as urban 
physicians (see discussion below on shipping). Its possible that in the east, with its shorter 
distances between rural areas and urban areas, that there is no urban/ rural price differential 
and that in other areas, there is a urban/ rural price differential. 

The preceding discussion indicates that there is mixed evidence for both regional and 
urban/ rural price variation. This evidence, however, is based strictly on conversations with 
distributors. Further, the magnitude of the price differentials, if they exist, is not known. 
With regard to how variation in local costs of distributors affects prices charged by 
distributors to physicians, no hard data is available. Thus, the existence of geographic input 
price variation for medical supplies and equipment cannot be verified from available sources. 



gpd\pracexp\sem\chap2 



III-2-5 



Shipping Costs 

Costs of shipping products from manufacturers to wholesalers and distributors affect 
the prices charged by local distributors. Since manufacturers are scattered over the entire 
country, 3 it is unlikely that distributors in any one Census Region incur higher shipping costs 
than distributors in the other Census Regions. We were concerned, however, whether the 
prices paid by physicians located in rural areas were higher than those paid by physicians 
located in the nearest urban area because of higher shipping costs. Finally, because Alaska, 
Hawaii, and Puerto Rico are very distant from the mainland, it is quite likely that shipping 
costs increase the acquisition costs of inputs. To determine the impact of shipping costs, 
then, on the prices paid by physicians in rural areas, and off-shore locales, we contacted 
several shipping firms as well as local distributors. 

Local distributors do not tend to charge different prices to urban and rural physicians 
for the same products because of transportation costs. That is, within their service or market 
area, they have one price (exclusive of volume discounts) for any given product. 4 Neither 
Federal Express (FedEx) nor United Parcel Service (UPS) charge extra for delivery to rural 
areas. FedEx, for instance, charges a flat rate of $52 to send a 40-pound package (second day 
economy service) from Boston to any business address on the U.S. mainland. Similarly, UPS 
charges $70.75 to ship a 70-pound package (second day air) from Boston to any business 
address on the U.S. mainland. While UPS does charge differential rates for ground 
transportation from Boston to other parts of the country, there is no urban/ rural difference 
within a zone. Thus, a 70-pound package sent from Boston to anywhere in eastern Nebraska 
(zone 6 from Boston) would cost $17.68 regardless of whether it was sent to Omaha or Grand 
Island. Larger shipments (both physically and in value) usually are shipped via national 
trucking companies rather than by FedEx or UPS. National trucking companies such as 
Consolidated Freightways charge on the basis of delivery to local trucking terminals (even if 
the delivery is actually made to a business address). Like UPS, Consolidated Freightways 
charges the same rate regardless of whether the delivery is made to an urban or to a rural 
location within the same "zone." 

Shipping costs to Hawaii, Alaska, and Puerto Rico are higher than to mainland 
destinations. For 40-pound package sent by second day air from Boston via FedEx, it would 



There may be concentrations of manufacturers in some areas (e.g., pharmaceutical manufacturers in Puerto Rico and CT- 
scans, MRIs, and X-ray machines in Ohio). 

4 Note, however, the preceding discussion regarding differences in competition among distributors with regard to urban 
and rural physicians. 



gpci\ pracexp\ sem\ chap2 



III-2-6 



cost an extra $5 to go Puerto Rico and an extra $10 to go to Hawaii or Alaska over the $52 
that it would cost to a mainland destination. For UPS, the situation is a little more 
complicated because of the ground/air options. To send a 70-pound package from Boston to 
Miami by second day air via UPS would cost $70.75 whereas it would cost $75.50 to go to 
Puerto Rico. Similarly, a package from Boston to the west coast would cost $70.75 whereas 
to go to Hawaii and "urban" Alaska would ccst $84.00 and to rural Alaska would cost 
$92.75. 5 Because ground transportation is an option for mainland destinations, ground air 
freight differentials to non-mainland destinations potentially understate the shipping 
differential to Alaska, Hawaii, and Puerto Rico. As was the case for ground transportation, 
larger shipments are handled by carriers other than FedEx and UPS. We contacted Emery 
Worldwide, a well-known air shipper of large items. To send a 500-pound item from 
Cleveland to Honolulu or Anchorage would cost about $1,400 while to San Francisco, it 
would cost about $1,100 - a differential of 27 percent. The differential (for the same 500- 
pound shipment) between Miami and Puerto Rico is 14 percent. 6 Even larger shipments to 
non-mainland destinations would be sent by a combination of trucking and ocean ships. 
Since shipping by ocean ships is cheaper than by air, the overall shipping cost differential 
from the west coast to Alaska and Hawaii is probably less than indicated by Emery's rate 
differentials. Our best guesses at the overall shipping cost differentials are 25 percent for 
Alaska and Hawaii and 15 percent for Puerto Pijo. Shipping costs, however, are only part of 
the overall acquisition costs of equipment and supplies. Thus, the shipping cost differentials 
that might be incorporated into price indices for medical supplies and medical equipment 
would be lower than the preceding estimates. 

Miscellaneous 

The amount of geographic price variation of items in the miscellaneous category 
varies. For instance, the prices of national journals and memberships in national professional 
associations do not have any geographical variation. Legal services, accounting services, and 



The differentials of 18.7 percent for "urban" Alaska and Hawaii, 6.7 percent for Puerto Rico, and 31.1 percent for rural 
Alaska are for a 70-pound item only. The percentage differentials are higher at lower weights. On average, the differential 
declines by .623 percentage points for each additional pound The inverse relationship between the percentage differential and 
weight, however, is not completely linear nor completely consistent (see Figure 2-1). 

Crates and other shipped objects that take up more space (higher volume), for a given weight, are charged higher rates. 



gpci\ pracexp\ sem\ chap2 



III-2-7 



1 p 



FIGURE 2-1 UPS RATE DIFFERENTIALS BETWEEN MAINLAND AND OFF-SHORE 
DESTINATIONS (2nd DAY AIR), FEBRUARY 1992 



UPS 2nu Day Air Off-Shore Rate Differentials 



200.0 -r 



180.0 -L "\ 




III-2-8 



computer programmer expenses have a large labor component that probably can be 
measured by the salaries of such professionals. Professional salaries can be obtained from 
the 1990 Census and can be aggregated to nearly any desired geographic definition. 
Professional automobile is a difficult category to analyze because of the amount of individual 
bargaining before a final sales price is agreed upon and the difficulty of specifying a 
representative automobile. Some information on geographic differences in automobile costs 
may be purchased from Runzheimer, International, located in Milwaukee, Wisconsin. 

The price or cost of continuing education (CE) also can vary considerably. Local 
hospitals often provide free CE to staff physicians during grand rounds. CE mail courses 
exist and their prices do not geographically vary. Some hospitals charge for intensive one or 
two-day courses. Non-local physicians would not only have to pay the course fee but also 
incur transportation and lodging expenses. Finally, CE is annually rotated so that one region 
of the country does not always have the lowest transportation costs. The costs for this type 
of CE will not vary geographically (except, perhaps, for an urban-rural transportation 
differential). Determining the typical CE mix of physicians and the "prices" paid is very 
difficult. 

2.1.3 Conclusions 

The following are the major conclusions we reached from our search for information 
on supplies, equipment, and miscellaneous input prices: 

1. There is currently no data available that is adequate to determine geographic 
variation in medical supplies, equipment, and miscellaneous input prices. The 
same conclusion was reached in a recent Urban Institute repc/t to HCFA 
(Zuckerman et al. 1992). The only databases that may warrant further 
consideration are the IMS America Hospital Supply Index and pharmaceutical 
databases. 

2. The anecdotal/ interview evidence on geographic variation in supplies and 
equipment prices is conflicting. Generally, it appears that manufacturers 
charge uniform prices, aside from shipping costs. Distributors' (wholesalers') 
prices may vary because of the local cost of doing business and competitive 
conditions. Larger physician groups are likely to be able to buy through 
national distributors with uniform prices, again aside from shipping costs. Solo 
practitioners and smaD group physicians are more likely to buy from local 
distributors with varying prices. Larger groups are also likely to receive 
volume discounts. 



gpci\pracexp\sem\chap2 



III-2-9 



3. Shipping costs to Alaska, Hawaii, and Puerto Rico are higher than to mainland 
United States destinations. However, shipping costs to rural areas do not 
appear to be higher than to nearby urban areas. 

2.2 Primary Data Collection 

Among the problems of secondary data on input prices is lack of comprehensiveness. 
That is, price information may be available only for a few inputs. Another problem is that 
there is no hard data, only impressions and/or anecdotes of uncertain validity. Further, the 
extreme heterogeneity of the "miscellaneous" expense category makes it difficult to obtain 
reliable secondary data. One way of obtaining more comprehensive price data is to conduct 
a survey. This section reviews basic survey design issues and discusses the relative merits of 
surveying physicians and equipment and supply firms. 

2.2.1 Design Issues for a Survey of Input Prices 

Given the hundreds, if not thousands, of types of medical supplies and equipment 
used in physician offices, it is necessary to select representative inputs for a survey. One 
major criteria for selection is that the medical supply or equipment is common to many 
practices. Another is that the item is believed to exhibit the same price variation as other 
inputs. It would probably be necessary to convene a panel of physician experts to specify the 
supplies and equipment that should be in the survey. To accommodate the distinctive 
equipment of specialists, 7 the survey's sample size needs to be very large in order to obtain 
statistical power. The overall sample size also has to be large enough so that the appropriate 
geographical categories have enough observations to support the construction of a 
geographic price index. This means that MSAs should not be the geographic level of the 
survey because of the large number physicians that would have to be surveyed. At the other 
extreme, using the four census regions would likely result in considerable intra-region 
variation. One possibility is to use the nine census divisions and, within each census 
division, define areas as rural, small MSA, and large MSA. Such a strategy would result in 
27 areas to be sampled. Given the necessity of obtaining sufficient specialist respondents, 27 
areas is probably the best comprise between accuracy and costliness of the survey. 



For instance, ophthalmologists have YAG laser machines, photo slit lamps, and automated visual filed test equipment 
machines whereas most internists and family practitioners do not use such specialized equipment (Center for Research in 
Ambulatory Health Care Adrninistration, 1991). 



gpci\pracexp\sem\chap2 



III-2-10 



Once representative supplies and equipment ha\ e been specified, it is also necessary 
to obtain information on unit prices (instead of the total amount spent on the items) and on 
the characteristics of the product. For instance, bandages may be 2-ply or 6-ply while non- 
adherent dressings may be non-medicated wet, medicated wet, or dry. It is even more 
important for equipment that product characteristics are obtained because rapid technological 
change can dramatically affect prices. For instance, CT-scanners, which used to be only 
housed in hospitals, are now present in some physician practices. For clinical chemistry 
testing equipment, the ability to perform tests on, for instance, blood uric acid, blood urea, 
serum cholesterol, and glucose need to be taken into account as well as whether the 
equipment performs automated multi-channel tests. Because of price inflation and because 
some changes of attributes may not be easily specified, it is important to know the year in 
which the equipment was purchased. Because of technological change, prices tend to be 
discounted on used equipment, thus data on whether the equipment was purchased new or 
used needs to be obtained. For instance, Abbott sold Vision 1, a chemistry analyzer, in 
1989/ 90 for $15,000 new. Today, because of technological change, a three year old used 
machine sells for about $1,500 even though the machine still has plenty of physical life left in 
it. Finally, the costs of maintenance contracts need to be factored in because the price for 
some equipment includes maintenance. 

While many if not most physicians purchase rather than rent/ lease equipment, a non- 
trivial number of physicians do rent or lease equipment (see Table 3-1). In addition to the 
basic rental/ lease price, other terms of the lease agreement such as its term and whether 
maintenance is included need to be obtained. One attraction of lease and rental agreement 
prices is that annualized values can be readily obtained. Having annualized values would be 
useful because medical equipment has a lifespan of several years. Thus, from an analytical 
viewpoint, leased and rented equipment prices do not need to be standardized as much as 
purchase prices. Fewer data adjustments would likely produce more reliable estimates of 
geographic input price variation. 

If the survey is to provide values for the construction of an actual index to 
geographically adjust payments, then cost shares for the medical supplies, equipment, and 
miscellaneous categories need to be developed. Two ways to develop cost shares are: (1) 
rely on existing survey of physician expense data such as PPCIS and SMS; or (2) a new 
survey of physicians that obtains cost share information in addition to unit prices. 



gpci\ pracexp\ sem\ chap2 



III-2-11 



2.2.2 Types of Primary Surveys 

Survey of Physicians 

A survey of physicians would offer the following advantages. First, actual transaction 
prices (including shipping costs) instead of list prices would be obtained. Second, the prices 
would be actual equipment purchased rather than hypothetical equipment. Third, price 
information on items in the miscellaneous category can be obtained. A survey of physicians 
would have some problems. First, physicians may not be able to accurately recall the 
purchase/ rental cost or the year in which the equipment was purchased. Similarly, they may 
not remember if the equipment was purchased new or used. Third, they may not know the 
relevant attributes of the equipment/ supplies. Fourth, a survey of physicians would be quite 
expensive. It would be more expensive than HCFA's Physician Practice Cost and Income 
Survey (PPCIS) because more physicians need to be surveyed, more types of supplies and 
equipment would have to be covered than by the PPCIS, and more questions would have to 
be asked regarding each input than in the PPCIS. Thus a primary survey of physician would 
cost considerably more than the 1988 PPCIS (which cost more than 2.8 million dollars). 

Survey of Suppliers 

One problem with surveying physicians is the large sample size needed to obtain 
enough representative physicians in the major specialties. This suggests the possibility of a 
survey of suppliers. The problem here is that physicians buy from a large variety of 
suppliers: local distributors, regional/ national distributors, and, for major equipment, 
directly from manufacturers. This means that the survey design may have to be quite 
complex. The identification of the supplier could also be quite difficult. Aside from major 
manufacturers such as General Electric that are readily identifiable, it will quite costly to 
identify local and regional/ national distributors. Other problems may include the 
unwillingness of suppliers to divulge what they regard as proprietary data. Even if they are 
willing to participate, they may not want to take the effort to provide actual transaction data, 
but rather just list prices. 

2.3 Concluding Comments 

Given the lack of reliable input price data, the existence and/ or extent of actual 
geographic input price variation can only be verified by a comprehensive survey of 



gpci\pracexp\sem\chap2 



III-2-12 



physicians or suppliers. A survey of physicians would be preferable to a survey of suppliers. 
The reasons are as follows. First, HCFA and other organizations have considerable 
experience in surveying physicians but little expeiience in surveying suppliers. Second, the 
design of a physician survey would be less complicated than would a survey of suppliers 
because of a readily available sampling frame (the AMA Masterfile). 



gpti\pracexp\sem\chap2 



III-2-13 



3.0 



ANALYSIS OF PHYSICIAN SURVEY DATA 



In this chapter we analyze survey data collected from physician practices to determine 
what it reveals about geographic variation in the prices of supplies, equipment, and 
miscellaneous inputs. Our data source is the 1988 HCFA Physician Practice Costs and 
Income Survey (PPCIS). Several similar analyses have been completed previously using 
American Medical Association (AMA) Socioeconomic Monitoring System (SMS) data. We 
begin by reviewing the results of these studies, in Section 3.1. We also discuss in this section 
the limitations of using expenditure data to infer price variation. Section 3.2 then presents 
our analysis of the PPCIS data. The PPCIS contains practice expenditures for supplies, 
equipment, and miscellaneous, similar to the SMS. In addition, however, an equipment 
supplement to the PPCIS collected data on purchase prices of several types of equipment 
used in physician practices. The PPCIS equipment supplement is a unique source of data 
that we use to study geographic variation in equipment prices. Subsections of Section 3.2 
discuss the PPCIS survey and data, our analytical methods, results on geographic 
expenditure/ price variation, and correlations of expenditure/ price variation with other GPCI 
price indices. 

3.1 Review of Previous Studies 
3.1.1 AMA Study 

Two previous studies have used AMA SMS data to examine geographic variation in 
physician expenditures and input prices. Researchers at the AMA (Gillis, Reynolds, and 
WiUke, 1992) used SMS data to analyze the validity of GPCI price indices. Annual data from 
1984 to 1991 was used in the analysis of physician supply expenditures. To approximate a 
price for medical supplies, reported supplies expenses were divided by an estimate of annual 
patient visits, under the assumption that supplies are a variable expense directly related to 
the number of visits. (It is not clear whether office or total patient visits were used as the 
denominator.) This "supplies price" was then regressed on dummy variables for Medicare 
payment localities and physician specialty. Gillis et al. found that the supplies price proxy 
varied significantly (in a statistical sense) across localities. They also tested the correlation of 
the supplies price proxy with the GPCI office rent, nonphysician employee wage, malpractice 
insurance, and physician work price indices. The nonphysician employee wage was the most 
highly correlated with the supplies price proxy, with an estimated elasticity of the supplies 



gpci\ pracexp\sem\chap3 



III-3-1 



price with respect to the wage index of .33. However, Gillis et al. considered the supply 
price proxy to be "especially weak" because of the lack of a good quantity measure of supply 
inputs. Thus, their analysis of it was limited to testing for geographic variation and 
correlation with other GPCI price indices. Gillis et al. did not analyze equipment or 
miscellaneous expenses because of the lack of a good quantity input measure to construct a 
price proxy. Their conclusion was that the apparent variation of supply prices across 
localities suggests that a price proxy may need to be constructed for supplies expenses. 

3.1.2 Urban Institute Study 

The Urban Institute recently submitted a report to HCFA that reexamined the price 
indices used in the GPCI (Zuckerman, Miller, Wade, and Pauly, 1992). Part of the evaluation 
involved analysis of 1990 and 1991 SMS expense data for supplies, equipment, and "other" 
inputs, and for the sum of these categories. The dataset the AMA made available to the 
Urban Institute did not include measures of physician outputs, such as visits. Therefore, 
prices were approximated by dividing expenses by the number of non-physician personnel. 

Expenses per nonphysician personnel were regressed on a set of payment locality 
(some localities were aggregated by the AMA for confidentiality reasons) and specialty 
dummy variables. The null hypothesis of no geographic variation (among localities) was 
rejected for all expense categories. Zuckerman et al. suggest that this implies that the current 
lack of any variation in the supplies, equipment, and miscellaneous components of the GPCI 
should be reconsidered. 

In addition, expenses per employee were regressed on various permutations of the 
GPCI office rental and nonphysician employee wage indices. In these regressions, the 
nonphysician employee wage index (based on 1980 Census wage data) had the largest 
elasticities. For the sum of supplies, equipment, and miscellaneous, and for equipment, the 
elasticity was not significantly different from 1.0, implying a proportional relationship. For 
supplies, the elasticity was significantly less than 1.0, and for miscellaneous, it was 
significantly greater than 1.0. 

Zuckerman et al. suggest that an "ad hoc" price adjustor for supplies, equipment, and 
miscellaneous (individually or summed) could be created from the GPCI nonphysician wage 
or office rental indices. The adjustment would be defined by the price index's coefficient 
(elasticity) in the regression of supplies, equipment, or miscellaneous expenses on it. For 
example, if the elasticity of Census wages in the double logarithmic regression of supplies 



gpci\ pracexp\ sem\ chap3 



III-3-2 



per employee on wages was .43, the adjustor would be the Census wage raised to the .43 
power. A similar approach was used in constructing the geographic capital cost adjustor in 
Medicare's Prospective Payment System for hospitals from the PPS area wage index. 

How to choose among the various possible "ad hoc" adjustors is not entirely clear 
because none of them have much face validity. One basis on which to choose would be 
percentage of variation in expenses explained, but Zuckerman et al. do not report the R- 
squares from their regressions. They seem to favor ad hoc adjustors that have a nearly 
proportional relationship to expenses. It is not clear why this should be a criterion, however, 
because the relationship can always be made proportional by transforming the independent 
variable (e.g., raising it to the power of its expense elasticity). 

3.1.3 Limitations of Using Expenditure Data to Infer Input Price Variation 

After reviewing these two previous studies, but before presenting our own analysis of 
physician survey data, it is useful to recognize the limitations of using expenditure data to 
infer input price variation. By definition, expenditures E for a practice input are the product 
of price P and quantity Q: 

E = PQ(P, X). (1) 

Quantity purchased is a function of price and othei factors X. From equation (1), it is clear 
that expenditures varies directly with price only if quantity is fixed. That is, expenditures is 
a good measure of price variation only if physician demand for an input is totally inelastic 
with respect to price. If demand is unit elastic, expenditures will be constant even if price 
varies, and thus will yield no information about price variation. If demand is unit elastic, a 
10 percent higher price would be offset by 10 percent lower quantity. If demand is elastic, 
expenditures will actually vary inversely with price. For example, prices might be lower in 
urban areas, but, if their demand is elastic, urban doctors would spend more on the input. 
Thus, expenditure variation does not necessarily reveal anything about price variation: 
expenditures may vary directly, not at all, or inversely with price. 

Other factors X besides price may also affect quantity of inputs purchased and thus 
weaken any relationship between price and expenditures. If these factors are uncorrected 
with geography (which is used as a proxy for the unmeasured price), the price /expenditure 
relationship will be weakened, but not biased. However, there is reason to believe that some 



gpd\praeexp\sem\chap3 



III-3-3 



factors that affect quantity are correlated with geography. For example, the types of patients 
seen could differ by area. Sicker patients may be more likely to travel to urban areas to be 
nearer to more extensive medical facilities. Therefore, urban physicians may see a 
systematically more difficult casemix than rural physicians, and thus use a greater quantity of 
supplies, equipment, and other inputs. Urban areas may also have more substitutes and 
complements (e.g., hospitals, ambulatory surgery centers, free-standing imaging centers) to 
physician services available, so physicians may stock fewer supplies in their offices. On the 
other hand, urban physicians may have more "high-tech" resources to work with. In 
practicing a more aggresive style of medicine, they may use more supplies and equipment. 
Regional differences may also exist. Hospital length of stay is shorter in the west, so western 
physicians may see sicker patients in their offices and use more resources. Thus, the 
coefficients of urban-rural, regional, and payment locality in regressions of expenditures on 
geography may reflect variations in quantity, not in price. 

In a regression equation, of course, "control" variables can be added to attempt to 
hold constant the quantities of inputs used, and thus isolate price variation. However, many 
of the factors affecting quantity are difficult to measure. Moreover, the same "control" 
variables may be correlated with both quantity and price. It then becomes very difficult to 
distinguish price from quantity variation. Even controlling for specialty, which is usually 
seen as the most basic factor to hold constant in an expenditure regression, can introduce 
biases. Specialists are more common in urban areas, so physician specialty is correlated with 
any urban-rural differences in input prices. Controlling for specialty may inadvertently 
reduce or eliminate the urban-rural price difference the analyst intends to measure in an 
expenditure regression. 

The best situation occurs when a good measure of quantity of the input is available 
on the physician survey. For example, square feet of office space was collected on both the 
1988 PPCIS and recent SMS surveys and can be used as a quantity measure for office costs 
(cost per square foot). The AMA researchers (Gillis et al, 1992) used visits as a quantity 
measure for supplies, but considered it to be inadequate. They did not analyze equipment or 
miscellaneous inputs at all because of lack of a quantity measure. The Urban Institute 
researchers (Zuckerman et al., 1992) were forced to use nonphysician employees as the 
quantity measure for all their inputs. Employees is clearly a crude approximation to the 
desired quantity measures. For this reason, the results of the Urban Institute report must be 
interpreted cautiously. 



gpci\ pracexp\sem\chap3 



III-3-4 



In short, analysis of expenditure data may reveal little or nothing about price 
variation. However, because almost no price data are available, expenditure data have been 
analyzed by us and other researchers. Results of these analyses must be interpreted 
cautiously and compared with a priori expectations and other information. Maintained 
assumptions (e.g., about the price elasticity of demand) should be kept in mind. Nevertheless, 
in the absence of price data, analysis of expenditures may be useful if interpreted properly. 

3.2 Analysis of PPCIS Practice Costs and Equipment Prices 

This section consists of four subsections. The first is a brief description of the PPCIS, 
the source of the data used in the analyses. Second, construction of analytical variables is 
discussed. Third, the results of the univariate and multivariate analysis are presented. 
Fourth, the use of current GPCI price indices as proxies to geographically adjust payments 
for medical supplies, medical equipment, and miscellaneous expenses is considered. 

3.2.1 Description of the 1988 Physicians' Practice Costs and Income Survey 

The data used in the descriptive and econometric analyses are from the 1988 
Physicians' Practice Costs and Income Survey (PPCTS). This was a telephone survey of 
physicians conducted by the National Opinion Research Center (NORC) for HCFA. The 
sample universe was nonfederal physicians who spent at least 20 hours per week in patient 
care and had completed their residencies. All specialties and group sizes were sampled, but 
physicians employed by hospitals, clinics, HMOs, and faculty practice plans were excluded. 
That is, only self-employed physicians or physicians employed by another physician or group 
of physicians were surveyed. The weighted response rate to the overall survey was 61 
percent. Questions regarding practice expenses for medical supplies, medical equipment 
(depreciation, interest, leases, and rentals), and miscellaneous expenses (including 
professional automobile and continuing education) were part of the main survey. 

HCFA developed an equipment supplement to the PPCIS concerning physicians' 
ownership, rental, and lease arrangements as well as investments in selected types of medical 
equipment. The equipment supplement was divided into two main categories: (1) diagnostic 
medical equipment and (2) laboratory equipment. The diagnostic testing section gathered 
purchase prices of 12 types of diagnostic equipment, including those requiring an expensive 
piece of equipment like magnetic resonance imaging (MRI), or a widely diffused and less 



gpci\ pracexp\ sem\ chap3 



III-3-5 



expensive device such as a sigmoidoscope. Physicians in all specialties were asked for the 
purchase price of three common types of diagnostic equipment and their use: X-ray, routine 
EKG, and diagnostic ultrasound. Questions on the remaining equipment were specialty- 
specific (see Table 3-1). Physicians in all specialties were asked about all four types of 
laboratory equipment. 



3.2.2 Methods 



Neither the practice expenses variables nor the equipment "price" variables were 
suitable for analysis without (separate) standardization procedures. The reason these 
variables needed to be standardized is because the objective of the analysis is to determine 
geographic price variation. The practice expense variables are the result of unit prices times 
the number units of input purchased. 1 The equipment price variables are a mixture of prices 
for single pieces of equipment and expenditures on multiple pieces of equipment. 2 

There are three types of practice expense variables: medical supplies, medical 
equipment, and miscellaneous. Since the units of inputs of each of these variables were not 
requested on the PPCIS, 3 another method of standardization was developed. Medical supply 
expenses, after being divided by the practice's FTE physicians, were standardized by the 
responding physician's annualized number of office visits. This assumes that medical 
supplies are physically consumed in each office visit. That is, medical supplies are variable 
inputs. Medical equipment expenses and miscellaneous expends, however, are relatively 
fixed inputs. While patient volume may affect the physical deterioration of medical 
equipment and business equipment, the allowable tax write-off is based on straight-line 
depreciation. The only "standardization" for medic M equipment and miscellaneou jxpenses 
were in the multivariate analysis where the number of full-time equivalent physicians was 
included as an independent variable to control for the overall size of the practice. 



'Even the depreciation and interest expense parts of the medical equipment and miscellaneous categories can be 
conceptualized in terms of price times quantity. 

2 

In the equipment supplement of the PPCIS, the purchase price question was worded as follows: "What [was the total 
purchase price/are the annual (lease/ rental) costs] of all the (INSERT) equipment currently (owned/ leased/ rented) by your 
practice?" "INSERT' refers to the type of equipment (e.g., X-ray) for which the question was asked. The word "all" in the 
questions means that the combined value of all equipment of a given type would be given as an answer. 

Vhich would have been difficult given the heterogeneous nature of the inputs within each of the three practice expense 
categories. 



gpd\pracexp\sem\chap3 



III-3-6 



TABLE 3-1 

NUMBER OF VALID OBSERVATIONS IN PPC'S EQUIPMENT SUPPLEMENT 



Rented 

Equipment Purchased or Leased 
Diagnostic 

X-Ray a 748 131 

EKG a 1 ,021 98 

Diagnostic Ultrasound a 473 130 

Upper Gl Endosopy b 105 19 

Flexible Sigmoidoscopy b 390 31 

Diagnostic Colonoscopy b 105 13 

Cardiovascular Stress Test c 89 18 

ECG c 142 63 

ECG d 60 10 

Cardiovascular Stress Test d 64 12 

Myocardium d 115 

CT Scan e 119 

MRI e 10 4 

Nuclear Scan e 7 2 

Laboratory a 

Clinical Chemistry 576 137 

Hematology 642 108 

Microbiology 400 39 

Histology 70 1 1 



All physicians. 

b Internists, general and family practitioners, general surgeons, and gastroenterologists only. 
c Internists, general and family practitioners only. 
° Cardiologists only. 
e Radiologists only. 



III-3-7 



gpci\pracexp\sem\TAB3-1 .XLS 



Medical equipment purchase "prices" were standardized two different ways. First, 
they were divided by the number of fun-time physicians in the practice and, second, by the 
annualized number of tests performed by each type of equipment. The number of 
observations available for analysis limited the types of equipment analyzed. The reason is 
that 100 observations are too few when analyzing geographic price variation, even over the 
four census regions. Thus equipment purchase prices analyzed are limited to X-ray, 
electrocardiographic monitoring (EKG), diagnostic ultrasound, flexible sigmoidoscopy, clinical 
chemistry, hematology, and microbiology (see Table 3-1). Further, after data editing, we 
were not able to analyze the rental or leasing of equipment because of too few observations; 
our anlaysis is limited to purchase prices. 4 

3.2.3 Geographic Variation in Practice Costs and Equipment Prices 

A descriptive (univariate) analysis was not performed on the three practice expense 
variables. The reason is that even after standardization for the number of physicians and/ or 
office visits, there would be considerable variation due to specialty mix. For instance, 
psychiatrists tend to have low medical supply and medical equipment costs relative to other 
specialties. Although there are between 1,776 and 2,663 over?U observations for the practice 
expense variables, with sixteen specialties there are not enough observations per specialty to 
support a specialty-specific analysis of geographic variation. 

Even after standardizing the equipment "prices," there was no control for the age 
(vintage) of the equipment, whether purchased used or new, equipment characteristics (e.g., 
ability to perform multi-channel tests), and multiple pieces of equipment. As one 
consequence of the lack of adequate controls, the equipment price distributions had many 
extreme values, to which mean prices were quite sensitive. Thus median prices, which are 
less influenced by extreme values than means, are presented in Table 3-2.° The medians 
sometimes indicate little variability while other instances, considerable variability. For X-ray 
machines and microbiology equipment, for instance, the purchase price per physician was 
identical in three of the four census regions. EKG machines had the same median in rural 



4 lhis loss was regrettable because the annualized value of rents and leases were available. The use of annualized rents and 
leases would have circumvented the problems posed by the multi-year lifespan of medical equipment. 

5 For meaningful comparison of prices in each area, medians only require that the median-priced equipment in each area 
has the same characteristics, not that the entire sample of equipment is comparable. 



gpci\pracexp\sem\chap3 



III-3-8 



TABLE 3-2 

MEDIAN EQUIPMENT PURCHASE PRICES BY CENSUS REGION AND URBANICITY 



CENSUS REGION 



URBANICITY 



MSA 



Type of Equipment 
X-Ray 

Purchase price per MD 

Purchase price per test 



ALL 



$10,492 
(502) 

21.59 
(454) 



Northeast 



$10,000 
(77) 



Midwest South 



Electrocardiographic Monitoring (EKG) 
Purchase price per MD 

Purchase price per test 

Diagnostic Ultrasound 
Purchase price per MD 



Purchase price per test 

Flexible Sigmoidoscopy 
Purchase price per MD 

Purchase price per test 

Clinical Chemistry 

Purchase price per MD 

Purchase price per test 

Hematology 

Purchase price per MD 

Purchase price per test 

Microbiology 

Purchase price per MD 

Purchase price per test 



1,667 
(689) 

6.92 
(630) 



10,000 
(363) 

67.31 
(318) 



2,350 
(271) 

25.64 
(239) 



3,333 
(393) 

2.88 
(341) 



1,464 
(412) 

1.92 
(363) 



500 
(237) 

1.10 
(215) 



20.43 
(64) 



1,500 
(146) 

5.77 
(136) 



12,750 
(56) 

79.33 
(50) 



2,500 
(44) 

20.60 
(41) 



2,766 
(61) 

3.61 
(49) 



1,050 
(66) 

1.54 
(56) 



300 

(37) 

0.58 
(31) 



$10,000 
(136) 

19.23 
(125) 



1,333 
(157) 

6.81 
(142) 



9,375 
(90) 

67.31 
(79) 



1,500 
(71) 

22.44 
(63) 



3,000 
(111) 

1.92 
(98) 



1,000 
(107) 

1.40 
(93) 



500 

(53) 

1.15 
(48) 



$14,100 
(193) 

25.64 
(179) 



1,875 
(243) 

7.21 
(227) 



11,786 
(139) 

70.83 
(126) 



1,975 
(88) 

25.00 
(77) 



4,000 
(159) 

2.88 
(140) 



2,417 
058) 

2.56 
(143) 



500 
(85) 

1.15 
(81) 



West 



$10,000 
(96) 

25.80 
(86) 



1,500 
(143) 

7.69 
(125) 



8,385 
(78) 

48.08 
(63) 



3,000 
(68) 

34.13 
(58) 



2,500 
(62) 

5.16 
(54) 



1,000 
(81) 

1.92 
(71) 



500 

(62) 

1.10 
(55) 



Rural 



$9,643 
(153) 

17.09 
(139) 



1,500 
(217) 

7.69 
(195) 



7,930 
(87) 

73.22 
(74) 



1,750 
(81) 

33.65 
(72) 



3,585 
(154) 

2.98 
(134) 



1,250 
(163) 

1.73 
(139) 



460 
(86) 

1.10 
(75) 



Small 



$12,000 
(237) 

24.26 
(212) 



1,500 
(291) 

6.92 
(263) 



11.500 
(184) 

64.10 

(162) 



2,000 
(119) 

20.19 
(104) 



3,333 
(159) 

2.64 
(138) 



1,350 
(170) 

1.54 
(154) 



500 
(100) 

1.38 
(94) 



Large 



$11,306 
(112) 

23.08 
(103) 



1,800 
(181) 

5.77 
(172) 



11,726 

(92) 

74.18 
(82) 



3,400 
(71) 

25.64 
(63) 



2,950 
(80) 

3.61 
(69) 



1,750 
(79) 

3.27 
(70) 



333 
(51) 

0.77 
(46) 



Note: Number of observations are in parentheses. 

Source: 1985 HCFA Physician Praclice Cost and Income Survey. 



III-3-9 



gpd\semrpf\TAB3-2.XLS 



areas and small urban areas. For other types of equipment, the medians reveal considerable 
geographic variation - 30-40 percent is not uncommon and there are several instances in 
which the highest median is about twice as large as the lowest median. The two methods of 
standardizing equipment prices give somewhat different results. Equipment prices 
standardized on the number of physicians indicate that input prices are often highest in the 
south and large urban areas and lowest in rural areas. Prices standardized on the number of 
tests indicate that input prices are often highest in the west and large urban areas, and lowest 
in the midwest, northeast, and small urban areas. Overall, though, there are no strong 
patterns, which is consistent with random variation due tc lack of controls for vintage and 
other characteristics. 

Part of the variation displayed by the medians in Table 3-2 could be due to 
characteristics that were not controlled for such as whether clinical chemistry equipment 
could perform multi-channel tests. Multivariate analysis was employed on all of the 
equipment price variables listed in Table 3-2 and on the three practice expense variables. For 
all regressions, the dependent variable was the natural logarithm of the expense or price 
variable. This means that if a physician reported zero expenses or zero price, the observation 
was excluded from analysis. This is appropriate given that the purpose of the study is to 
determine price variation. Independent variables common to all regressions were census 
region dummy variables (northeast is the omiU'-d region), urbanicity dummy variables (rural 
is the omitted category), and specialty (general/ family practice is the omitted specialty). 
Specialty variables are included because some specialties may buy/ lease more expensive 
machines than others, or use a different mix/quality of supplies and miscellaneous inputs. 7 
Some regressions had additional independent variables. The equipment and miscellaneous 
expense regressions included the natural logarithm of the number of fuU-time equivalent 
physicians in the practice to control for practice size. The diagnostic ultrasound, clinical 
chemistry, and hematology regressions included attributes such as the types of tests the 
machines could perform and whether the machines could perform automated multi-channel 
tests. Finally, outlier trims were applied to all regressions: the upper and lower 10 percent 
of equipment prices were dropped from analysis with less severe trims on the practice 



6 In no circumstance does one area have the majority of high °r l° w medians. 
7 

The control for specialty is probably most important with regards to machines for which we do not have information on 
special characteristics or on overall capacity. 

gpd\pracexp\sem\chap3 III-3-10 



expense variables. The magnitudes of geographic coefficients are sensitive to extreme values 
of the dependent variable - partly because of omitted (unavailable) qualitative controls. 

The results of the practice expense regressions are presented in Table 3-3. Because of 
the large number of equipment price regressions, only one, for EKG machines, is presented, 
in Table 3-4. The medical supplies expense per visit regression had the largest regression 
coefficients on the census region variables, and they suggest much lower expenses in the 
northeast. While all three of the census region variables were statistically significant in the 
supplies regressions, neither of the urbanicity variables were (and they also had much 
smaller regression coefficients). All three of the census region variables were statistically 
significant in the miscellaneous expense regression but had much smaller magnitudes than in 
the supplies regression. Again, neither urbanicity variable was significant in the 
miscellaneous regression and the magnitudes of their coefficients were smaller than the 
census region coefficients. Only one geographic dummy variable was statistically significant 
in the equipment regression - although the magnitudes of the geographic coefficients were 
similar to the miscellaneous expense geographic coefficients. Overall, seven of the nine 
census region regression coefficients are statistically significant in the three practice expense 
regressions but none of the urbanicity variables were statistically significant. One possible 
reason why all of the census region variables have positive signs is that physicians in the 
northeast may be more likely than physicians elsewhere to use hospitals or other nonoffice 
settings more intensively (e.g., have tests performed in hospitals rather than in their office). 

Unlike the practice expense regressions, the census region variables were rarely 
statistically significant in the equipment price regressions (not shown). Conversely, there 
were several statistically significant urbanicity variables, more than can be accounted for by 
chance alone. The signs on the urbanicity variables (rural was the omitted category) were 
usually positive, indicating rural physicians faced lower input prices than urban physicians. 
While the equipment regressions generally had very low adjusted R squares, there were 
several that had values .15 or higher. 

To determine whether there is statistically significant geographic variation in expenses 
and equipment purchase prices, F tests for the joint significance of the geographic variables 
were performed. The results of the F tests are in Table 3-5. In the first set of columns, the 
test is for the presence of regional differences, in the second set of columns the test is for 
differences by urbanicity, and in the final set of columns, the test is for either regional or 
urbanicity effects. 



gpci\pracexp\sem\chap3 



III-3-11 



TABLE 3-3 



PRACTICE EXPENSE REGRESSIONS 





Morlif*£il Q 1 i rtnlioc 


MiQppll^inAni ac 


C«| U 1 |JI 1 Id 1 1 




Pynpncp Ppt \/iQit 
uAMviioc rci v ion 


pynpnQp 




Intercept 


-0.400 *** 


-0.994 *** 


-1.038 *** 




(0.094) 


(0.062) 


(0.124) 


Midwest 


0.257 *** 


0.146 *** 


0.144 




(0.081) 


(0.051) 


(0.104) 


South 


0.333 *** 


0.100 ** 


0.163 * 




(0 074) 


(0 047) 


(0.095) 


West 


0.254 *** 


0.138 *** 


0.139 




(0.080) 


(0.051) 


(0.102) 


Omnll MO A 

omaii MoA 


a ao 4 

0.031 


0.050 


0.052 




(0.066) 


(0.042) 


(0.082) 


Large MoA 


0.078 


-0.002 


A A A 4 

-0.101 




(0.075) 


(0.048) 


(0.098) 


Internal Medicine 


0.060 


0.089 


0.1 1 1 




(0.096) 


(0.067) 


(0.126) 


Cardiology 


0.290 


0.272 


0.473 *** 




(0.140) 


(0.098) 


(0.183) 


Gastroenterology 


n one ** 


n oco *** 






(0.136) 


(0.091 ) 


(0.176) 


vjiner ivieascai 


U.z/O 


U.U / o 


-U. 1 ID 




(0.109) 


(0.074) 


(0.144) 


\jvi iviai vjui Uui y 


081 


201 *** 

U.£U 1 


-0.085 




(0.114) 


(0.076) 


(0.150) 


Orthopedic Surgery 


0.398 *** 


0.444 *** 


0.492 *** 




(0.136) 


(0.093) 


(0 171) 


Oohthalmoloav 


-0.189 


0.485 *** 


0.931 *** 




(0 1 24) 


(0.090) 


(0 163) 


Uroloav 


0.464 *** 


0.166 ** 


0.172 




(0.119) 


(0.081) 


(0.158) 


Obstetrics/Gynecology 


0.246 


0.209 


A AA4 
0.091 




(0.108) 


(0.075) 


(0.144) 


Cardiovascular Surgery 


O *l OA 

0.109 


0.362 


A CO A *** 




(0.142) 


(0.095) 


(0.195) 


Other Surgery 


0.277 


0.313 


A A AO *** 




(0.118) 


(0.084) 


(0.156) 


Psychiatry 


-0.750 


-0.1 15 


A OOO *** 




(0.160) 


(0.082) 


(0.216) 


Anesthesiology 


— 


0.032 


A O AO *** 

-0.893 






(0.084) 


(0.227) 


Radiology 


— -— 


0.179 


A OAA ** 

0.399 










. .... 

Other Specialities 


O.lbo 


A. * Oft ** 
U. PJO 






-(0.185) 


(0.095) 


(0.209) 


Log FTE MDs 




1.002 *** 


0.992 *** 






(0.017) 


(0.032) 


Observations 


1754 


2436 


1614 


Adjusted R Squared 


0.047 


0.624 


0.410 


F 


5.772 *** 


193.499 *** 


54.436 *** 



'Indicates statistical significance at the 10 percent level. 
"Indicates statistical significance at the 5 percent level. 

-Indicates statistical significance at the 1 percent level. gpci\pracexp\sem\TAB3-3.XLS 



NOTE: Standard errors are in parentheses. Equation estimated by ordinary least squares. 



TABLE 3-4 

ELECTROCARDIOLOGY EQUIPMENT (EKG) PRICE REGRESSIONS 





EKG Price 


EKG Price 




oer Test 


per MD 


Intercept 


0.113 


-0.078 




(0.121) 


(0.105) 


Midwest 


-0.117 


-0.093 




(0.120) 


(0.105) 


South 


-0.136 


0.102 




(0.108) 


(0.093) 


West 


0.029 


0.054 




(0.119) 


(0.103) 


Small MSA 


0.017 


-0.144 * 




(0.096) 


(0.084) 


Large MSA 


-0.027 


0.004 




(0.110) 


(0.096) 


Internal Medicine 


-0.208 ** 


0.173 " 




(0.101) 


(0.088) 


Cardiology 


-0.124 


0.475 *■ 




(0.132) 


(0.120) 


Gastroenterology 


0.097 


-0.105 




(0.165) 


(0.139) 


Other Medical 


0.098 


0.051 




(0.162) 


(0.136) 


General Surgery 


0.271 


0.107 




(0.228) 


(0.192) 


Orthopedic Surgery 


0.252 


-0.169 




(0.505) 


(0.399) 


Ophthalmology 


0.425 


0.883 * 




(0.312) 


(0.397) 


Urology 


0.664 


-0.338 




(0.440) 


(0.329) 


Obstetrics/Gynecology 


0.062 


-0.064 




(0.312) 


(0.234) 


Cardiovascular Surgery 


-0.337 


-0.130 




(0.236) 


(0.208) 


Other Surgery 


— 


0.866 






(0.560) 


Anesthesiology 


0.818 ** 


0.595 * 




(0.394) 


(0.357) 


Radiology 


0.223 


-0.501 




(C.501) 


(0.557) 


Other Specialities 


0.037 


-0.049 




(0.394) 


(0.360) 


Observations 


509 


556 


Adjusted R Squared 


.018 


.049 


F 


1.508 * 


2.493 * 



'Indicates statistical significance at the 10 percent level. 
"Indicates statistical significance at the 5 percent level. 
'"Indicates statistical significance at the 1 percent level. 

NOTE: Steward en'ors are in parentheses. Equation estimate d by ordinary least squares. 



III-3-13 



gpci\pracexp\sem\TAB3-4.XLS 



0) 

o 
c 

(0 

o 
c 
c 
o 

o5 



x— OO 00 

O T" 

O O t- 

d ci d 



00 00 

CD CD 

T- O 

ci d 



r- t- 

o m 

t- CD 

O O 



O CN 

00 00 

o> o 

o o 



o o 

O T- 

d o 



cn o 
co o 

ci ci 



CN tJ- 

T- O 

o o 

ci d 



m 

h- CO 

CO CN 

ci ci 



o 

'■3 

w 

'■3 

2 

(0 



in o) t 

ID CN CO 

CN I — CD 

^ CN 



r- co 

co r- 

m o 

^ CN 



Oi CO 

r- CD 

00 CD 

■r^ ci 



O) CN 
CO 00 
CN OJ 



O CO 

oo r*~ 

CO ID 

CN t- 



O CO 
Tj- in 

o o 



cn t- 

cn T~ 

cn cd 

CN CO 



CN 

h~ oo 

O CO 



(0 
111 
_1 
03 
< 
Of 
< 
> 

O 
X 

2 

O 
O 
UJ 

o 



o 
c 

CO 

n 



Oi 
u 
c 
re 

o 

E 
c 
o 

<75 



CO 

re 
+-> 

(0 



CO 00 00 

t- oo 

CO CO t— 

odd 



r~~ co 

m n 

ID t- CD 

O' T-' 



O 00 

CD t- 

CN CO 

ci O 



T- O 

CO t- 



oo m 

o o 

t- 0> 

d d 



o> o 

co o 

cn -e 

cn d 



cn h» 

CN CM 

cn o 

d d 



t- CD 

OO CD 

O CD 

d co 



■>* oo 

CN CD 

T- O 

d d 



CM t- 

T- CO 

CM CM 



CO CN 

CN CD 

r- CD 

d d 



CD CJ) 

CN CD 

CO CO 

d d 



cn 
co o 
o 

d d 



CD oo 

co m 

oo oo 

d 



CM CD 

o o 
d d 



in co 

T- O 

CO CD 

CN d 



c 
o 

'5> 

4) 

(A 
3 
(A 
C 

a> 
O 



0} 

o 
c 
re 
o 
|3 

c 
g 



.52 

+3 

re 
*■> 

(0 



O h- 

o t- in 
o o co 

odd 



oo cn cn 

CO r- N 

O Tl" o 

N (O t- 



m cn 

cn is) 

o o 

d d 



o 

t- CD 

CN CM 



t- CD 
CN CO 

d d 



T- CD 

m o 



cn co 

CM 00 

h- CO 

d d 



co m 
co cn 

d d 



CM t- 

oo co 

r- CD 

d d 



r- 

co h- 
co m 

d 



cn o 

o o 

co o 

d d 



-<3- oo 

o Tj- 
CM CD 



in 
o cn 
o o 

d d 



CM 

o ■<* 

CD t- 
CM 



CM CO 

cn cn 

in o 

d d 



co r-- 

m o 

co -^r 

d CM 



CO 
CO 

£ 

O 

a> 

0£ 



CO 
> 

8 

E 
o 

Q co 

2 1- c to 
g <D 0) W 
<0 Q. Q. (/) 

x: 3 ~ o- 
0. (/) S LU 



u 



10 

re 



3 
0. 

C 

a> 

E 
a 

'5 
cr 

LU 

re 

o 

t5 
a> 
S 



Q T7> 
3 <P 



a> 
Q. 

0) 
O 

•c 

CL 
03 



q: or 



Q to 
cu 



CD 
CL 



O 



a> 

CL 

8 

•c 

Cl 



O 
XL * 
LU LU 



co 



CD 
CL 

Oi 

o 
■c 

CL 

■a 
c 

o 

CO 
03 



CD 
CL 



CL 

-a 



a io 

t_ i_ 

CD CD 

Cl Cl 

8 8 

c c 

Cl Cl 

>. >. 

Cl Cl 

O O 



a 
co 
o 
■g 
o 
£ 



o 
co 
o 
■g 
'o 
E 



D Z) 



(7) tV) 



Q w 

CD CD 
Cl Cl 

CD CD 
O O 

q. a. 

to 
E 'E 

CD CD 

o o 

re re 
g g 

c c 
O O 



CD CD 
Cl Cl 

<D 0> 

g g 

Q. Q. 

CO CD 

g g 
o o 



03 

E 

a> 
I 



03 

E 

CD 

X 



CD 



CD CD 
CL Cl 

CD CD 

g g 

a. o. 
>> 

CD CD 

g g 
g g 
la 1q 

2 2 
o g 

^ 5 



III-3-14 



Among the practice expense regressions, the census region variables were jointly 
significant in the supplies and miscellaneous regressions but not the equipment regression. 
The urbanicity variables were not jointly significant in any of the three practice expense 
regressions. As might be expected by the previous discussion of the results, all geographic 
variables were significant in only the supplies and miscellaneous regressions. 

Regional effects were significant in only two of the 14 equipment price regressions 
(clinical chemistry per test and hematology per MD). Urbanicity effects were significant in 
only two of the regressions (ultrasound per test and hematology per test). For all geographic 
effects, only four of the 14 regressions had significant locational effects. 

Overall, the results of Table 3-5 provide little consistent evidence of geographic input 
price variation. However, the number of statistically significant coefficients, in twelve out of 
51 regressions, is too high to ascribe to chance alone. 

The univariate and multivariate analyses provide weak evidence that geographic 
input price variation might exist. The low number of statistically significant geographic 
effects, the inconsistency of results, and the poor data quality do not provide confidence that 
the statistical results provide strong evidence of geographic input price variation. Although 
the evidence for geographic input price variation is weak, it cannot be ruled out either. 
Better input price data obtained through primary data collection would be necessary to 
establish more definitively the degree of geographic price variation. 

3.2.4 Correlation with Other GPCI Input Price Measures 

In lieu of obtaining better, but expensive, supplies, equipment, and miscellaneous 
input price data, existing price indices for other inputs may be adequate to adjust physician 
payments. In place of the geographic dummy variables used above, the employee wage 
component of the current GPCI, the office rent component of the current GPCI, and the 
overall GPCI are alternatively used in the expenditure/ price regressions. The natural 
logarithm of each of the GPCI variables was used; thus, the regression coefficients can be 
interpreted as elasticities. An elasticity of one indicates that a one percent increase in the 
GPCI is associated with a one percent increase in practice expense or equipment price. 

The results of using GPCI components as independent variables in lieu of geographic 
variables are presented in Table 3-6. The positive signs of the GPCI price indices in the 
medical equipment purchase price regressions are inconsistent with the negative signs of the 
GPCI price indices in the physician expense regressions. The regression coefficients for the 



gpd\pracexp\sem\chap3 



III-3-15 



TABLE 3-6 



ELASTICITIES OF GPCI PRICE INDICES IN EXPENSE/PRICE REGRESSIONS 



GPCI PRICE INDEX 



Rearession 


Non-Physician 
Employee Waqes 


Office Rent 


Overall 


Physician Expenses 

Supply expenses per MD office visit 
Miscellaneous Expenses 
Equipment Expenses 


-0.239 
-0.063 
-0.558 * 


-0.053 
-0.061 
-0.195 


-0.156 ** 

-0.14b 

-0.359 


rvieaicai equipment rurcnase rrice 

X-Ray price per MD 
X-Ray price per test 


0.037 
0.387 


-0.058 
0.091 


0.118 
0.145 


EKG price per MD 
EKG price per test 


0.034 
-0.047 


0.014 
0.119 


0.205 
0.198 


Ultrasound price per MD 
Ultrasound price per test 


-0.066 
0.848 * 


-0.149 
0.121 


-0.027 
0.634 


Sigmoidoscopy price per MD 
Sigmoidoscopy price per test 


0.831 * 
0.033 


0.465 ** 
-0.009 


1.106 ** 
0.106 


Clinical chemistry price per MD 
Clinical chemistry price per test 


-0.636 
0.725 


-0.223 
0.5 7 3 ** 


-0.652 
0.763 


Hematology price per MD 
Hematology price per test 


0.156 
1 .855 *** 


0.204 
0.979 *** 


0.352 
1.637 *** 


Microbiology price per MD 
Microbiology price per test 


-0.524 
-1.108 


-0.003 
-0.468 


-0.268 
-0.936 



'Indicates statistical significance at the 10 percent level. 
"Indicates statistical significance at the 5 percent level. 
"'Indicates statistical significance at the 1 percent level. 



III-3-16 



gpci\pracexp\sem\lab3-6 



GPCI indices in our practice expense regressions are always negative but only statistically 
significant twice. In contrast, in the AMA and Urban Institute's practice expense regressions, 
the GPCI indices had positive signs. Thus, our findings are inconsistent with the AMA and 
Urban Institute findings, which are both based on AMA data. 

The GPCI employee wage index is significant in three of the 14 equipment purchase 
price regressions (ultrasound per test, sigmoidoscopy per MD, and hematology per test). The 
office rental index is also significant in three of the 14 equipment regressions (with clinical 
chemistry per test instead of ultrasound per test). The overall GPCI is significant only in the 
sigmoidoscopy per MD and the hematology per test equipment purchase price regressions. 

Overall, ten of the 51 practice expense and equipment purchase price regression GPCI 
coefficients are statistically significant - too high to be due to chance alone, but not strong 
enough to indicate that the GPCI components are adequate price indices for supplies, 
equipment, and miscellaneous prices. 

3.2.5 Conclusions 

This section analyzed geographic variation in supply, equipment, and miscellaneous 
expenses collected on HCFA's 1988 Physicians' Practice Costs and Income Survey (PPCIS). 
All analyses controlled for physician specialty. Supply expenses were standardized by an 
estimate of annual office visits. In analyzing equipment and miscellaneous expenses, we 
controlled only for the number of fuU-time equivalent physicians in the practice. Our major 
conclusions are: 

(1) None of these three expenses vary significantly by urbanicity (rural, small 
metropolitan, large metropolitan). 

(2) Supply and miscellaneous expenses are significantly lower in the northeast 
than in the other three Census regions. 

(3) The overall GPCI and the nonphysician employee wage and office rental 
indices are generally not significantly related to supplies, equipment, and 
miscellaneous expenses. To the extent they are related, the correlations are 
negative . This fmding differs markedly from the positive, significant 
relationship found by the Urban Institute with AMA data (Zuckerman, et al, 
1992). The PPCIS results suggest that expenses are higher in rural and small 
cities than in large cities. 



gpti\pracexp\sem\chap>3 



III-3-17 



We also analyzed PPCIS data on diagnostic and laboratory equipment purchase prices 
per FTE physician and per test performed. Seven types of equipment had sufficient sample 
sizes to be analyzed: X-ray, EKG, ultrasound, sigmoidoscopy, clinical chemistry, hematology, 
and microbiology. We found some weak evidence of geographic variation in these prices, 
but the patterns of variation were not consistent. Our results could be explained by random 
variation in the data due to our inability to control for equipment vintage and characteristics. 
However, we cannot rule out geographic price variation with ihese data. Unlike expenses, 
some equipment prices were significantly positively related to the GPCI price indices such as 
nonphysician wages. 



gpci\ pracexp\ sem\ cha p3 



III-3-18 



4.0 CONCLUSIONS FOR POLICY 



Given the uncertainty over actual geographic variation in supplies, equipment, and 
miscellaneous prices, before discussing policy options, it is useful to consider the effect of 
hypothethical price variation in supplies, equipment, and miscellaneous inputs on the overall 
Medicare Fee Schedule Geographic Adjustment Factor (GAF). These simulations are 
presented in Section 4.1. The conclusion of the simulations is that under any reasonable 
scenario for price variation, the inaccuracy in the GAF from not making an adjustment for 
supplies, equipment, and miscellaneous prices is very small. Policy options for the supplies, 
equipment, and miscellaneous component of the updated (1996) practice expense GPCI are 
discussed in Section 4.2. Section 4.3 concludes the report with some brief comments on 
strategies for primary data collection in the longer run. 

4.1 Effect of Hypothetical Supplies, Equipment, and Miscellaneous Price Variation on the 
Geographic Adjustment Factor (GAF) 

The effect of price variation in any of its components on the GAF 1 depends on the 
share of the input in the GAF as well as the extent of price differences. Table 4-1 shows the 
GAF market basket shares of supplies, equipment, miscellaneous inputs, hypothetical input 
price variation unweighted and weighted by each input's market basket share, and the 
highest and lowest index values implied by the hypothetical price variation. For example, 
medical supplies has a 5 percent share in the GAF. Suppose supplies prices varied from 10 
percent above to 10 percent below the national average. Weighted by supplies' market 
basket share, the impact on the overall GAF is only from 0.5 percent above to 0.5 percent 
below the national average, or implies a highest index value of 1.005 and a lowest index 
value of 0.995. 

Table 4-1 shows that hypothetical price variation in the supplies, equipment, and 
miscellaneous inputs probably does not have a large effect on the GAF. Even 20 percent 
price variation in the sum of the three categories implies only a 2.7 percent difference from 
the national average when weighted by the market basket share. The lack of impact is 
because of the small market basket shares of these inputs. Equipment has an especially small 



By GAF, we refer to the overall Geographic Adjustment Factor utilized in the Medicare Fee Schedule, which is a weighted 
sum of the physician work, practice expense, and malpractice GPQs. 



gpd\pracexp\sem\chap4 



III-4-1 



TABLE 4-1 



EFFECTS ON GAF OF HYPOTHETICAL VARIATION IN SUPPLIES, EQUIPMENT, 
MISCELLANEOUS INPUT PRICES 



High/Low 







Hypothetical 


Weighted 


Index Values 




Market 


Price 


by Market 


Weighted 


GAF Component 


Basket Share 1 


Variation 2 


Basket Share 


bv Share 3 


Supplies 


5.0 


±10 % 


±0.5 % 


.995 


1.005 






±20 


±1.0 


.990 


1.010 


Equipment 


2.4 


±10 


±0.2 


.998 


1.002 






±20 


±0.5 


.995 


1.005 


Miscellaneous 


6.0 


±10 


±0.6 


.994 


1.006 






±20 


±1.2 


.988 


1.012 


Total 


13.4 


±10 


±1.3 


.987 


1.013 






±20 


±2.7 


.973 


1.027 



1 Market basket shares are in the current (1992) GAF. 
2 Relative to national average price. 
3 1 .0 is the national average. 

NOTE: GAr" is Medicare Fee Schedule Geographic Adjustment Factor. 



III-4-2 



gpci\pracexp\sem\TAB4-1 .XLS 



share, at 2.4 percent. Of course, if the actual degree of price variation is greater than 
assumed in Table 4-1, the effect on the GAF would be larger. However, large price variation 
in the continental United States seems unlikely. Supplies and equipment are products and 
arbitrage (i.e., transshipment of products in response to regional price differences) should 
limit geographic price differences unless shipping costs are very high. Some elements in the 
miscellaneous category (e.g., legal and accounting services) are labor-related and could 
exhibit greater variation. However, other miscellaneous elements (e.g., office furniture) are 
nonlabor inputs whose prices may not vary much geographically. Also, geographic price 
variation in the various miscellaneous inputs may not be highly correlated and thus, the 
overall price variation of the miscellaneous category may be limited. The small impact of the 
supplies, equipment, and miscellaneous categories on the GAF limits any inequities that may 
result from lack of an adjustment for these prices. Also, the relative unimportance of these 
categories should be considered when evaluating the desirability of expensive efforts to 
collect primary data on these input prices. 

4.2 Policy Options 

There is currently no conclusive evidence of geographic variation in the prices of 
supplies, equipment, and miscellaneous inputs employed by physician practices. A priori 
considerations and anecdotal evidence suggest that any price variation is small for most areas 
and inputs. Analyses of expenditure data suggest that variation may exist, but they are not 
entirely convincing due to the difficulty of inferring price variation irom expenditure 
variation. No input price data exists that is adequate for making an adjustment for supplies, 
equipment, and miscellaneous prices in the GPCI. 

Given this background, the short run policy options are: 

• no adjustment; 

• no adjustment for the mainland United States, but an outlier adjustment (for 
supplies and equipment only) for Alaska, Hawaii, and Puerto Rico based on 
higher shipping costs; or 

• an "ad hoc" adjustment based on the coefficient of an available price index 
(e.g., the GPCI nonphysician employee wage index) in a regression explaining 
physician supplies, equipment, and/ or miscellaneous expenses. 



gpd\pracexp\sem\chap>4 



III-4-3 



The first option maintains the status quo. As shown by the simulations in Section 4.1, any 
inaccuracy in the GAF resulting from lack of an adjustment is likely to be very minor. The 
second option would establish an add-on for outlying areas (paid for by a slight reduction in 
payments to all other areas), justified by higher shipping costs. This option is consistent with 
current policy in Medicare's Prospective Payment System (PPS) for hospitals, where an add- 
on for hospital nonlabor expenses is made to outlying regions. However, there is no 
evidence that supplies, equipment, and miscellaneous input prices of physician practices are 
actually higher in outlying areas. For example, many pharmaceuticals are manufactured in 
Puerto Rico, so shipping costs for drugs are presumably lower than to the mainland United 
States. The third option would base an adjustment on a regression of physician practice 
expenses, gathered from surveys, on available price indices. This method is consistent with 
the geographic adjustment for hospital capital costs in PPS, which is based on a regression of 
capital expenses on the PPS hospital wage index. However, the face validity of this 
procedure is limited, and it is difficult empirically to distinguish price variation from 
expenditure variation (expenditures are affected by quantities in addition to prices). 
Moreover, physician expenditure regressions using AMA SMS and HCFA PPCIS data are 
inconsistent. 

With these considerations in mind, the Health Care Financing Administration decided 
to maintain the policy of no adjustment for supplies, equipment, and miscellaneous prices in 
the updated (1996) practice expense GPCI. 

4.3 Primary Data Collection in the Long Run 

Currently, no data are available to determin. the extent of geographic vari£ ..jn in 
supplies, equipment, and miscellaneous prices. To further analyze price variation, or to 
develop a geographic adjustment, primary data collection is necessary. Although it may be 
feasible to collect such data, doing so would be expensive. Furthermore, the improvements 
in the accuracy of the GAF resulting from primary data collection are likely to be small, both 
because of the small share of supplies, equipment, and miscellaneous inputs in the GAF, and 
because their price variation is likely to be limited. The analysis presented in Section 4.1 
indicates that the likely improvement in the GAF's accuracy for the highest and lowest priced 
areas would be 1 to 3 percent. For most areas, the improvement would be much smaller. 
Given competing budgetary priorities, it is a policy decision for HCFA whether such gains 
justify the cost necessary to achieve them. Gains in the accuracy of the GAF per dollar spent 



gpd\pracexp\sem\chap4 



III-4-4 



may be greater for other analyses or data collection efforts. If HCFA does decide to mount a 
data collection effort, we believe that a survey of physicianr is preferrable to a survey of 
suppliers or distributors. 



gpci\pracexp\sem\chap4 



III-4-5 



REFERENCES 



Gillis, K. D., R. A. Reynolds, and R. J. Willke. Assessing the Validity of the Geographic 
Practice Cost Indexes, American Medical Association, September 23, 1992. 

Medical Economics, Volume 69, No. 2 (November 16, 1992). 

Pope, G., K. Adamache, and J. Webb, Geographic Variation in Hospital Nonlabor Input Prices 
and Expenditures, Draft Final Report, HCFA Cooperative Agreement No. 99-C-98526/1-08, 
August 1992. 

Welch, W.P., S. Zuckerman, and G. Pope, The Geographic Medicare Economic Index: 
Alternative Approaches, Final Report, HCFA Cooperative Agreement No. 17-C-98758/1-03, 
June 1989. 

Zuckerman, S., Assessment of the Accuracy of the Overhead GPCI, Draft Final Report, HCFA 
Contract No. 500-89-0054/3, November 1992. 



gpci\ pracexp\sem\ refs 



R-1 



PART IV: 
THE PRACTICE EXPENSE GPCI 



PART IV: THE PRACTICE EXPENSE GPCI 



The 1996 practice expense GPCI is the weighted sum of its three component indices- 
employee wages; office rents; and supplies, equipment, and miscellaneous prices-according 
to the following formula: 

practice expense GPCI = 0.398*(employee wage index) + 0.251*(office rental index) + 
0.351*(supplies, equipment, miscellaneous price index). 1 

In first three parts of this report, we identified the best options for updating the employee 
wage index, the office rental index, and the supplies, miscellaneous, and equipment price 
indices, respectively. To recapitulate: 

• the updated (1996) employee wage index is based on the median hourly 
earnings of occupations employed in physicians' offices, as measured by a 
special tabulation of the 1990 Census; 

• the updated office rental index is derived from the fiscal year 1994 Fair Market 
Rents published by the Department of Housing and Urban Development in the 
April 6, 1994 Federal Register; and 

• the updated supplies, equipment, and miscellaneous price index is 1.000 
everywhere, that is, no geographic adjustment is made for these prices. 

The 1996 practice expense GPCI is weighted by practice expense relative value units or RVUs 
(rather than population), reflects the Office of Management and Budget's June 1993 
metropolitan area redefinitions, 2 and captures county-specific wage and rent variation within 
large metropolitan areas (CMS As). These are all features— in addition to its use of more 
recent input price data— that distinguish the updated (1996) practice expense GPCI from the 
current (1992) practice expense GPCI. Of these changes, the use of more recent input price 



'The weights are the average shares of each of the three cost categories— employee wages; office rent; and supplies, 
equipment, and miscellaneous— in physician practice expenses. Cost c hares for the updated (1996) GPCI are discussed in detail 
in a companion report (Dayhoff, Schneider, and Pope, 1994). 



Although the practice expense GPCI is ultimately computed for Medicare payment localities, Census and HUD tabulate 
the underlying wage and rental data for metropolitan areas and the nonmetropolitan parts of states before it is crosswalked to 
payment localities. 



gpci\ pracexp\ parti v 



IV-1 



localities. Appendix 1V-1 shows that weighting the practice expense GPCI by RVUs versus 
population has only a minor effect on the index values of Medicare payment localities. 

Appendix IV-2 displays the 1996 practice expense GPCI, the 1996 rescaled practice 
expense GPCI, the 1995 rescaled practice expense GPCI, and the 1992 practice expense GPCI 
by Medicare payment locality. HCFA actuaries developed a multiplicative factor (1.00125) 
that was used to ensure that total Medicare Fee Schedule payments for practice expense are 
not affected by the updated practice expense GPCI. The 1996 rescaled GPCI reflects this 
factor. This rescaled GPCI will be used for payment beginning in 1996. The 1995 rescaled 
GPCI is the simple average of the 1996 rescaled GPCI and the 1992 GPCI. The 1995 rescaled 
GPCI will be used for payment in 1995 only. 

Appendix rV-3 ranks the Medicare payment localities in order of descending 
difference between the 1996 rescaled practice expense GPCI and the 1992 practice expense 
GPCI. The changes in the practice expense GPCI index values from updating (1996 versus 
1992 GPCI) are moderate. The largest increase is 12 percent, in Southwest Connecticut. Only 
three payment localities gain more than 10 percent. The largest loss is also 12 percent, in 
Peoria, Illinois. Only three localities lose more than 10 percent. The practice expense GPCI 
comprises 41 percent of the overall Medicare Fee Schedule Geographic Adjustment Factor 
(GAF). Therefore, Medicare fees for a typical service will change by less than 5 percent in 
any area due to updating the practice expense GPCI. Moreover, Congress has mandated a 
two-year transition from the 1992 to the 1996 GPCIs. The change in the 1995 transitional 
GPCI is only half as large as the already small difference between the 1996 and 1992 GPCIs. 



gpci\ pracexp\ partiv 



IV-2 



CHS LIBRARY 




3 aDT5 00D1H1D0 a